[\u6587\u732eSCI-FS-EN-2510281]PlantCaFo: An efficient few-shot plant disease recognition method based on foundation models<\/a><\/blockquote><\/iframe>\n<\/div><\/figure>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity is-style-dots\"\/>\n\n\n\n<h2 class=\"wp-block-heading has-text-align-center\">1.\u7814\u7a76\u80cc\u666f\u4e0e\u95ee\u9898<\/h2>\n\n\n\n\u8fd9\u7bc7\u8bba\u6587\u7684\u5f15\u8a00\uff08Introduction\uff09\u91cc\u63d0\u5230\uff0c\u81ea\u52a8\u8bc6\u522b\u690d\u7269\u75c5\u5bb3\u5bf9\u4e8e\u7cae\u98df\u5b89\u5168\u548c\u63d0\u9ad8\u4ea7\u91cf\u975e\u5e38\u91cd\u8981 \u3002\u867d\u7136\u73b0\u5728\u7684\u65b9\u6cd5\uff08\u6bd4\u5982\u5927\u578b\u795e\u7ecf\u7f51\u7edc\uff09\u53d6\u5f97\u4e86\u4e00\u4e9b\u8fdb\u5c55\uff0c\u4f46\u5b83\u4eec\u975e\u5e38\u4f9d\u8d56\u5927\u91cf\u3001\u6709\u6807\u7b7e\u7684\u6570\u636e \u3002<\/p>\n\n\n\n\u8fd9\u5728\u519c\u4e1a\u9886\u57df\u662f\u4e00\u4e2a\u5de8\u5927\u7684\u6311\u6218\u3002\u8fd9\u7bc7\u8bba\u6587\u660e\u786e\u6307\u51fa\u4e86\u4e24\u4e2a\u4e3b\u8981\u95ee\u9898\uff1a<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n\u3010\u4e3b\u8981\u95ee\u98981.\u6570\u636e\u74f6\u9888\u3011<\/p>\n\n\n\nHowever, this reliance presents significant challenges in agriculture. One challenge is that the collection and annotation of agricultural data are often expensive and time-intensive. <\/p>\n\n\n\n\u3010\u4e3b\u8981\u95ee\u98982.\u6837\u672c\u7a00\u6709\u6027\u3011<\/p>\n\n\n\nFurthermore, the rarity of certain plant diseases makes gathering a large number of examples impractical.<\/p>\n<\/blockquote>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u6570\u636e\u74f6\u9888\uff1a<\/strong>\u6536\u96c6\u548c\u6807\u6ce8\u519c\u4e1a\u6570\u636e\uff08\u6bd4\u5982\u75c5\u5bb3\u53f6\u7247\uff09\u901a\u5e38\u65e2\u6602\u8d35\u53c8\u8d39\u65f6 \u3002<\/li>\n\n\n\n<li>\u6837\u672c\u7a00\u6709\u6027\uff1a<\/strong>\u67d0\u4e9b\u690d\u7269\u75c5\u5bb3\u975e\u5e38\u7f55\u89c1\uff0c\u8fd9\u4f7f\u5f97\u6536\u96c6\u5927\u91cf\u6837\u672c\u53d8\u5f97\u4e0d\u5207\u5b9e\u9645 \u3002<\/li>\n<\/ol>\n\n\n\n\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u201c\u6570\u636e\u4f9d\u8d56\u201d\u7684\u74f6\u9888\uff0c\u7814\u7a76\u4eba\u5458\u8f6c\u5411\u4e86\u4e00\u79cd\u53eb\u505a\u201c\u5c11\u6837\u672c\u5b66\u4e60\u201d\uff08few-shot learning\uff09\u7684\u6280\u672f \u3002<\/p>\n\n\n\n\u5173\u4e8e\u8fd9\u4e2a\u80cc\u666f\uff0c\u6211\u4eec\u63a5\u4e0b\u6765\u53ef\u4ee5\u6df1\u5165\u63a2\u8ba8\u4e24\u4e2a\u65b9\u5411\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li class=\"has-medium-font-size\">\u4ec0\u4e48\u662f\u201c\u5c11\u6837\u672c\u5b66\u4e60\u201d\uff1f<\/strong>\u5177\u4f53\u4e86\u89e3\u4e00\u4e0b\u5b83\u662f\u5982\u4f55\u5de5\u4f5c\u7684\uff0c\u6bd4\u5982\u8bba\u6587\u4e2d\u63d0\u5230\u7684 \"N-way K-shot\" \u662f\u4ec0\u4e48\u610f\u601d \uff1f<\/li>\n\n\n\n<li class=\"has-medium-font-size\">\u73b0\u6709\u7684\u201c\u5c11\u6837\u672c\u5b66\u4e60\u201d\u65b9\u6cd5\u6709\u4ec0\u4e48\u95ee\u9898\uff1f<\/strong>\u4e3a\u4ec0\u4e48\u5b83\u4eec\uff08\u4f8b\u5982\u6570\u636e\u589e\u5f3a\u3001\u5143\u5b66\u4e60\u3001\u8fc1\u79fb\u5b66\u4e60\uff09\u8fd8\u4e0d\u591f\u597d\uff0c\u4ee5\u81f3\u4e8e\u9700\u8981\u8fd9\u7bc7\u8bba\u6587\u63d0\u51fa\u65b0\u6a21\u578b \uff1f<\/li>\n<\/ol>\n\n\n\n<h2 class=\"wp-block-heading\">1.1 \u4ec0\u4e48\u662f\u201c\u5c11\u6837\u672c\u5b66\u4e60\u201d<\/h2>\n\n\n\n\u6b63\u5982\u8bba\u6587\u6240\u8bf4\uff0c\u8fd9\u662f\u4e00\u79cd\u65e8\u5728\u4f7f\u7528\u201c\u5c11\u91cf\u6807\u8bb0\u6837\u672c\u201d\u6765\u8bad\u7ec3\u6a21\u578b\u7684\u6280\u672f \u3002<\/p>\n\n\n\n\u4e3a\u4e86\u5b9e\u73b0\u8fd9\u4e00\u70b9\uff0c\u5b83\u901a\u5e38\u4f1a\u7528\u5230\u4e24\u4e2a\u90e8\u5206\uff1a<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n\u3010\u652f\u6301\u96c6\u3011The support set contains a few labeled examples that the model uses to learn, \u3010\u67e5\u8be2\u96c6\u3011whereas the query set is used to evaluate the model's ability to generalize. <\/p>\n<\/blockquote>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u652f\u6301\u96c6 (Support set)\uff1a\u5305\u542b\u5c11\u91cf\u5e26\u6807\u7b7e\u7684\u6837\u672c\uff0c\u6a21\u578b\u7528\u5b83\u6765\u5b66\u4e60 \u3002<\/li>\n\n\n\n<li>\u67e5\u8be2\u96c6 (Query set)\uff1a\u7528\u6765\u8bc4\u4f30\u6a21\u578b\u5b66\u4e60\u540e\u7684\u6cdb\u5316\u80fd\u529b \u3002<\/li>\n<\/ul>\n\n\n\n\u4e3a\u4e86\u8bc4\u4f30\u8fd9\u79cd\u5b66\u4e60\u6548\u679c\uff0c\u8bba\u6587\u63d0\u5230\u4e86\u4e00\u4e2a\u5173\u952e\u6846\u67b6\uff0c\u53eb\u505a \"N-way K-shot\" <\/strong>\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6839\u636e\u8bba\u6587\u7684\u63cf\u8ff0 \uff0c\u4f60\u80fd\u8bd5\u7740\u89e3\u91ca\u4e00\u4e0b\u2018N-way\u2019\uff08N\u5143\uff09\u548c\u2018K-shot\u2019\uff08K\u6837\u672c\uff09\u5206\u522b\u4ee3\u8868\u4ec0\u4e48\u610f\u601d\u5417\uff1f<\/li>\n\n\n\n<li>\u201cN-way\u201d (N\u5143) \u6307\u7684\u5c31\u662f N \u4e2a\u4e0d\u540c\u7684\u7c7b\u522b\uff08\u6bd4\u5982 N \u79cd\u4e0d\u540c\u7684\u690d\u7269\u75c5\u5bb3\uff09\uff1bK-shot \u662f \u6bcf\u4e2a<\/strong> \u7c7b\u522b\u6709\u591a\u5c11\u4e2a\u6837\u672c\u3002<\/li>\n<\/ul>\n\n\n\n\u201c5-way\u201d\u6307\u7684\u662f5\u4e2a\u4e0d\u540c\u7684\u7c7b\u522b\uff08\u4f8b\u5982\u201c\u82f9\u679c\u9ed1\u661f\u75c5\u201d\u3001\u201c\u7389\u7c73\u9508\u75c5\u201d\u7b49\uff09\u3002\u201c1-shot\u201d\u6307\u7684\u662f\u6bcf\u4e2a\u7c7b\u522b\u53ea\u6709\u4e00\u4e2a\u5e26\u6807\u7b7e\u7684\u6837\u672c\u3002\u56e0\u6b64\uff0c\u201cN-way K-shot\u201d\u662f\u5c11\u6837\u672c\u5b66\u4e60\u7684\u6846\u67b6\uff0c\u5b83\u5b9a\u4e49\u4e86\u6a21\u578b\u9700\u8981\u5b66\u4e60\u7684\u7c7b\u522b\u6570\u91cf\uff08N\uff09\u4ee5\u53ca\u6bcf\u4e2a\u7c7b\u522b\u7684\u6837\u672c\u6570\u91cf\uff08K\uff09\u3002<\/p>\n\n\n\n\u73b0\u5728\u6211\u4eec\u7406\u89e3\u4e86\u95ee\u9898\uff08\u9700\u8981\u4ece\u5c11\u91cf\u6837\u672c\u4e2d\u5b66\u4e60\uff09\uff0c\u63a5\u4e0b\u6765\u8ba9\u6211\u4eec\u770b\u770b\u73b0\u6709\u7684\u89e3\u51b3\u65b9\u6848\u53ca\u5176\u5c40\u9650\u6027\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">1.2 \u73b0\u6709\u7684\u201c\u5c11\u6837\u672c\u5b66\u4e60\u201d\u65b9\u6cd5\u6709\u4ec0\u4e48\u95ee\u9898\uff1f<\/strong><\/h2>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nSignificant progress has been made in this area, primarily through three approaches: data augmentation<\/strong>[1], meta-learning<\/strong>[2] and transfer learning[3]<\/strong>. <\/p>\n<\/blockquote>\n\n\n\n\u8bba\u6587\u63d0\u5230\u4e86\u4e09\u79cd\u4e3b\u8981\u65b9\u6cd5\uff1a\u6570\u636e\u589e\u5f3a[1]\u3001\u5143\u5b66\u4e60[2]\u548c\u8fc1\u79fb\u5b66\u4e60[3]\u3002\u8bba\u6587\u6307\u51fa\u8fd9\u4e9b\u65b9\u6cd5\u5b58\u5728\u54ea\u4e9b\u6311\u6218\u6216\u95ee\u9898\uff0c\u5c24\u5176\u662f\u5728\u690d\u7269\u75c5\u5bb3\u65b9\u9762\uff1f<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nhowever, most of these methods require pretraining <\/strong>within the relevant domain.<\/p>\n\n\n\nIn recent years, few-shot learning based on transfer learning for plant disease classification has typically employed a two-stage strategy: first, learning general feature representations on a large number of relevant source sets<\/strong> and then fine-tuning on target sets to generate specific feature representations for subsequent prediction tasks.<\/p>\n\n\n\nHowever, these methods require a large amount of data and computational resources to train the feature extractor, and they often struggle with challenges such as class imbalance and domain shift, which hinder their generalization performance<\/strong>.<\/p>\n<\/blockquote>\n\n\n\n\u8fc1\u79fb\u5b66\u4e60\u7684\u5927\u591a\u6570\u65b9\u6cd5\u90fd\u8981\u8fdb\u884c\u4e0e\u8bad\u7ec3\uff0c\u5148\u5728\u5927\u91cf\u76f8\u5173\u7684\u6e90\u6570\u636e\u96c6\u4e0a\u5b66\u4e60\u901a\u7528\u7279\u5f81\u8868\u793a\uff0c\u7136\u540e\u518d\u5728\u76ee\u6807\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u5fae\u8c03\uff0c\u751f\u6210\u7528\u4e8e\u540e\u7eed\u9884\u6d4b\u4efb\u52a1\u7684\u7279\u5b9a\u7279\u5f81\u8868\u793a\u3002\u7136\u800c\u56e0\u4e3a\u8fd9\u4e9b\u65b9\u6cd5\u9700\u8981\u5927\u91cf\u7684\u6570\u636e\u548c\u8ba1\u7b97\u8d44\u6e90\u6765\u8bad\u7ec3\u7279\u5f81\u63d0\u53d6\u5668\uff0c\u5e76\u4e14\u5e38\u5e38\u9762\u4e34\u4e0d\u5e73\u8861\u548c\u9886\u57df\u504f\u79fb\u7684\u6311\u6218\uff0c\u4ece\u800c\u5f71\u54cd\u4e86\u5176\u6cdb\u5316\u6027\u80fd\u3002<\/p>\n\n\n\n\u8fd9\u6b63\u662f\u8bba\u6587\u6307\u51fa\u7684\u5173\u952e\u95ee\u9898\uff1a<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nHepsag et al. [21] proposed refining a model initially trained on ImageNet [22] with PlantCLEF2022 [23], which includes nearly 4 million images across 80,000 categories, to extract embeddings. They then trained a support vector machine, yielding an accuracy of 88.4 % in a \u201c38-way 10-shot\u201d scenario. <\/p>\n<\/blockquote>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u9700\u8981\u5927\u91cf\u6570\u636e\u548c\u8ba1\u7b97\u8d44\u6e90\uff1a<\/strong>\u4f20\u7edf\u7684\u8fc1\u79fb\u5b66\u4e60\uff08transfer learning\uff09\u901a\u5e38\u9700\u8981\u4e00\u4e2a\u5e9e\u5927\u7684\u3001\u76f8\u5173\u7684\u6e90\u6570\u636e\u96c6\uff08\u6bd4\u5982 PlantCLEF2022<\/strong>\uff0c\u4e00\u4e2a\u5305\u542b\u8fd1400\u4e07\u5f20\u56fe\u50cf\u7684\u6570\u636e\u96c6\uff09\u6765\u8fdb\u884c\u7b2c\u4e00\u9636\u6bb5\u7684\u9884\u8bad\u7ec3 \u3002<\/li>\n\n\n\n<li>\u6cdb\u5316\u6027\u80fd\u53d7\u9650\uff1a<\/strong>\u5b83\u4eec\u5e38\u5e38\u96be\u4ee5\u5e94\u5bf9\u7c7b\u522b\u4e0d\u5e73\u8861\u548c\u201c\u9886\u57df\u504f\u79fb\u201d\uff08domain shift\uff09\u2014\u2014 \u6bd4\u5982\uff0c\u5728\u5b9e\u9a8c\u5ba4\u62cd\u7684\u53f6\u5b50\u548c\u5728\u91ce\u5916\u62cd\u7684\u53f6\u5b50\u957f\u5f97\u4e0d\u4e00\u6837 \u3002<\/li>\n<\/ol>\n\n\n\n\u8fd9\u7bc7\u8bba\u6587\u6b63\u662f\u6293\u4f4f\u4e86\u8fd9\u4e2a\u75db\u70b9\uff0c\u63d0\u51fa\u4e86\u4e00\u4e2a\u66ff\u4ee3\u65b9\u6848\u3002<\/p>\n\n\n\n\u65e2\u7136\u4f20\u7edf\u7684\u8fc1\u79fb\u5b66\u4e60\u6709\u8fd9\u4e9b\u95ee\u9898\uff0c\u8bba\u6587\u662f\u53d7\u5230\u4e86\u4ec0\u4e48\u65b0\u6280\u672f\u7684\u542f\u53d1\uff0c\u4ece\u800c\u91c7\u7528\u4e86\u4e0d\u540c\u7684\u65b9\u6cd5\u5462\uff1f<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nInspired by the remarkable performance of foundation models such as CLIP [28] and DINO [29] in zero-shot and few-shot learning, we adopt existing large models to generate embeddings for samples in this work,thus alleviating the need for extensive data and limiting computational costs.<\/p>\n\n\n\n\u4f5c\u8005\u53d7\u5230\u4e86CLIP\u548cDINO\u7b49\u57fa\u7840\u6a21\u578b\u5728\u96f6\u6837\u672c\u548c\u5c11\u6837\u672c\u5b66\u4e60\u4e2d\u5353\u8d8a\u8868\u73b0\u7684\u542f\u53d1\uff0c\u7528\u5927\u6a21\u578b\u6765\u751f\u6210\u6837\u672c\u7684\u5d4c\u5165\uff0c\u4ece\u800c\u51cf\u8f7b\u4e86\u5bf9\u5927\u91cf\u6570\u636e\u7684\u9700\u6c42\u5e76\u9650\u5236\u4e86\u8ba1\u7b97\u6210\u672c\u3002<\/p>\n<\/blockquote>\n\n\n\n\u4f46\u8bba\u6587\u968f\u540e\u6307\u51fa\u4e86\u4e00\u4e2a\u95ee\u9898\u3002\u4f60\u4e0d\u80fd\u76f4\u63a5\u5c06\u50cfCLIP\u8fd9\u6837\u7684\u901a\u7528\u6a21\u578b\u201c\u5f00\u7bb1\u5373\u7528\u201d\u5730\u5e94\u7528\u4e8e\u50cf\u690d\u7269\u75c5\u5bb3\u8fd9\u6837\u9ad8\u5ea6\u5177\u4f53\u7684\u4efb\u52a1\uff0c\u5e76\u671f\u671b\u83b7\u5f97\u5b8c\u7f8e\u7684\u7ed3\u679c\u3002<\/p>\n\n\n\n\u8bba\u6587\u6307\u51fa\uff0c\u8fd9\u4e9b\u57fa\u7840\u6a21\u578b\u5728\u5e94\u7528\u4e8e\u519c\u4e1a\u9886\u57df\u65f6\u5b58\u5728\u54ea\u4e9b\u5c40\u9650\u6027\u6216\u6311\u6218\uff1f<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nHowever, existing foundation models have clear limitations in the agricultural field, such as mismatched datasets and poor generalization in agricultural scenarios,<\/strong> necessitating adjustments to address these issues.<\/p>\n<\/blockquote>\n\n\n\n\u8fd9\u7bc7\u8bba\u6587\u8981\u89e3\u51b3\u7684\u6838\u5fc3\u77db\u76fe\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u57fa\u7840\u6a21\u578b\uff08Foundation Models\uff09\u5f88\u5f3a\u5927\uff0c\u53ef\u4ee5\u5e2e\u6211\u4eec\u7701\u53bb\u9884\u8bad\u7ec3\u3002<\/li>\n\n\n\n<li>\u4f46\u5b83\u4eec\u65e2\u4e0d\u662f\u4e3a\u519c\u4e1a\u201c\u91cf\u8eab\u5b9a\u505a\u201d\u7684\uff08\u6570\u636e\u96c6\u4e0d\u5339\u914d\u3001\u6cdb\u5316\u80fd\u529b\u5dee\uff09\uff0c\u4e5f\u4e0d\u80fd\u76f4\u63a5\u201c\u66b4\u529b\u201d\u5fae\u8c03\uff08\u53c2\u6570\u592a\u591a\uff0c\u5bb9\u6613\u8fc7\u62df\u5408\uff09\u3002<\/li>\n<\/ol>\n\n\n\n\u8fd9\u5c31\u5f15\u51fa\u4e86\u4e00\u4e2a\u5173\u952e\u95ee\u9898\uff1a\u6211\u4eec\u5982\u4f55\u5728\u4e0d\u201c\u5b8c\u5168\u5fae\u8c03\u201d\u6574\u4e2a\u5e9e\u5927\u6a21\u578b\u7684\u524d\u63d0\u4e0b\uff0c\u8ba9\u5b83\u201c\u9002\u5e94\u201d\u6211\u4eec\u7684\u690d\u7269\u75c5\u5bb3\u8bc6\u522b\u4efb\u52a1\u5462\uff1f<\/p>\n\n\n\n\u8bba\u6587\u4e2d\u63d0\u5230\u4e86\u4e00\u4e2a\u89e3\u51b3\u8fd9\u7c7b\u95ee\u9898\u7684\u901a\u7528\u7b56\u7565\uff0c\u4f60\u77e5\u9053\u662f\u4ec0\u4e48\u5417\uff1f<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nTo address these challenges, several adapter-based methods have been proposed [32], which quickly adapt pretraining models to downstream tasks by introducing a few learnable parameters.<\/p>\n\n\n\nTrans:\u63d0\u51fa\u51e0\u79cd\u57fa\u4e8e\u9002\u914d\u5668\u7684\u65b9\u6cd5\uff0c\u901a\u8fc7\u5f15\u5165\u4e00\u4e9b\u53ef\u5b66\u4e60\u7684\u53c2\u6570\uff0c\u5feb\u901f\u5c06\u9884\u8bad\u7ec3\u6a21\u578b\u9002\u5e94\u4e0b\u6e38\u4efb\u52a1<\/p>\n<\/blockquote>\n\n\n\n\u8fd9\u5c31\u662f\u8bba\u6587\u63d0\u5230\u7684\u89e3\u51b3\u65b9\u6848\uff1a\u57fa\u4e8e\u9002\u914d\u5668\uff08adapter-based methods\uff09<\/strong>\u7684\u65b9\u6cd5 \u3002<\/p>\n\n\n\n\u8fd9\u79cd\u65b9\u6cd5\u975e\u5e38\u5de7\u5999\uff0c\u5b83\u4e0d\u662f\u53bb\u201c\u5b8c\u5168\u5fae\u8c03\u201d\uff08full fine-tuning\uff09\u90a3\u4e2a\u62e5\u6709\u4ebf\u4e07\u53c2\u6570\u7684\u57fa\u7840\u6a21\u578b\uff0c\u800c\u662f\u201c\u51bb\u7ed3\u201d\u57fa\u7840\u6a21\u578b\u7684\u7edd\u5927\u90e8\u5206\u53c2\u6570\uff0c\u53ea\u5f15\u5165\u4e00\u4e9b\u975e\u5e38\u5c11\u91cf\u7684\u3001\u53ef\u5b66\u4e60\u7684\u65b0\u53c2\u6570\uff08\u5373\u201c\u9002\u914d\u5668\u201d\uff09\uff0c\u8ba9\u6a21\u578b\u5feb\u901f\u9002\u5e94\u4e0b\u6e38\u7684\u65b0\u4efb\u52a1\uff08\u6bd4\u5982\u6211\u4eec\u7684\u690d\u7269\u75c5\u5bb3\u8bc6\u522b\uff09 \u3002<\/p>\n\n\n\n\u8fd9\u65e2\u5229\u7528\u4e86\u57fa\u7840\u6a21\u578b\u7684\u5f3a\u5927\u80fd\u529b\uff0c\u53c8\u907f\u514d\u4e86\u8fc7\u62df\u5408\u548c\u9ad8\u6602\u7684\u8bad\u7ec3\u6210\u672c\u3002<\/p>\n\n\n\n\u73b0\u5728\u6211\u4eec\u5df2\u7ecf\u6e05\u695a\u4e86\u7814\u7a76\u80cc\u666f\u548c\u6311\u6218\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u95ee\u9898\uff1a<\/strong>\u8bc6\u522b\u690d\u7269\u75c5\u5bb3\u7f3a\u4e4f\u5927\u91cf\u6570\u636e \u3002<\/li>\n\n\n\n<li>\u65b9\u5411\uff1a<\/strong>\u91c7\u7528\u201c\u5c11\u6837\u672c\u5b66\u4e60\u201d\uff08few-shot learning\uff09\u3002<\/li>\n\n\n\n<li>\u6311\u6218\uff1a<\/strong>\u4f20\u7edf\u5c11\u6837\u672c\u65b9\u6cd5\u9700\u8981\u9886\u57df\u9884\u8bad\u7ec3 \uff0c\u800c\u901a\u7528\u7684\u201c\u57fa\u7840\u6a21\u578b\u201d\uff08\u5982 CLIP\uff09\u76f4\u63a5\u7528\u6548\u679c\u4e0d\u597d\uff0c\u4e14\u5fae\u8c03\u56f0\u96be \u3002<\/li>\n\n\n\n<li>\u7b56\u7565\uff1a<\/strong>\u4f7f\u7528\u201c\u9002\u914d\u5668\u201d\uff08Adapter\uff09\u6765\u9ad8\u6548\u5730\u5fae\u8c03\u57fa\u7840\u6a21\u578b \u3002<\/li>\n<\/ul>\n\n\n\n\u63a5\u4e0b\u6765\uff0c\u6211\u4eec\u5c31\u5e94\u8be5\u805a\u7126\u4e8e\u8fd9\u7bc7\u8bba\u6587\u7684\u6838\u5fc3\u65b9\u6cd5\u4e86\uff1a<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity is-style-wide\"\/>\n\n\n\n<h2 class=\"wp-block-heading has-text-align-center\">2.\u6838\u5fc3\u65b9\u6cd5<\/h2>\n\n\n\n<div class=\"wp-block-group\"><div class=\"wp-block-group__inner-container is-layout-constrained wp-block-group-is-layout-constrained\">\n\u6838\u5fc3\u65b9\u6cd5 (PlantCaFo)\uff1a\u770b\u770b\u8fd9\u7bc7\u8bba\u6587\u5177\u4f53\u8bbe\u8ba1\u4e86\u4ec0\u4e48\u6837\u7684\u9002\u914d\u5668\u6765\u89e3\u51b3\u95ee\u9898\u3002\u8bba\u6587\u56fe1\u548c\u6458\u8981\u4e2d\u63d0\u5230\u4e86\u4e24\u4e2a\u5173\u952e\u6a21\u5757\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>DCon-Adapter (\u8f7b\u91cf\u7ea7\u6269\u5f20\u4e0a\u4e0b\u6587\u9002\u914d\u5668)<\/strong><\/li>\n\n\n\n<li>WDM (\u6743\u91cd\u5206\u89e3\u77e9\u9635)<\/strong><\/li>\n<\/ol>\n<\/div><\/div>\n\n\n\n<h2 class=\"wp-block-heading\">2.1 DCon-Adapter (\u8f7b\u91cf\u7ea7\u6269\u5f20\u4e0a\u4e0b\u6587\u9002\u914d\u5668)<\/strong><\/h2>\n\n\n\n\u8fd9\u662f\u4e00\u4e2a\u975e\u5e38\u5173\u952e\u7684\u6a21\u5757\u3002\u8fd8\u8bb0\u5f97\u6211\u4eec\u521a\u624d\u8ba8\u8bba\u8fc7\uff0c\u76f4\u63a5\u5fae\u8c03\u6574\u4e2a CLIP \u8fd9\u6837\u7684\u5927\u6a21\u578b\u5f88\u5bb9\u6613\u5728\u5c11\u6837\u672c\u6570\u636e\u4e0a\u201c\u8fc7\u62df\u5408\u201d (overfitting) \u5417\uff1f<\/p>\n\n\n\nDCon-Adapter \u5c31\u662f\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\u800c\u8bbe\u8ba1\u7684\u3002\u5b83\u662f\u4e00\u4e2a\u201c\u8f7b\u91cf\u7ea7\u201d\u7684\u6a21\u5757 \uff0c\u610f\u5473\u7740\u5b83\u53ea\u6709\u5f88\u5c11\u7684\u53c2\u6570\u9700\u8981\u8bad\u7ec3\u3002<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nIt consists of four layers: the first layer is a dilated convolution layer, which captures global features by expanding the receptive field, which is[\u3010\u597d\u5904\u3011particularly useful for handling complex backgrounds in plant disease recognition tasks<\/strong>]; the second layer is a batch normalization (BN) layer, which standardizes feature distributions to accelerate training and improve stability; the third layer uses the ReLU activation function, which introduces nonlinearity to enhance learning capacity and offers computational efficiency due to its simple derivative, accelerating the backpropagation process; and the fourth layer is a standard convolution layer, which is used to refine local features, further improving the model's classification ability in few-shot settings.<\/p>\n<\/blockquote>\n\n\n\n\u8fd9\u4e2a\u9002\u914d\u5668\u7531\u56db\u5c42\u7ed3\u6784\u7ec4\u6210 \u3002\u8bba\u6587\u7279\u522b\u5f3a\u8c03\u4e86\u5b83\u7684\u7b2c\u4e00\u5c42\uff0c\u5b83\u4e0d\u662f\u4e00\u4e2a\u666e\u901a\u7684\u5377\u79ef\u5c42\uff0c\u800c\u662f\u4e00\u4e2a\u201c\u6269\u5f20\u5377\u79ef\u201d (dilated convolution) \u5c42<\/strong> \u3002<\/p>\n\n\n\n\u6839\u636e\u8bba\u6587 2.4.1 \u8282\u7684\u63cf\u8ff0\uff0c\u4f7f\u7528\u8fd9\u79cd\u201c\u6269\u5f20\u5377\u79ef\u201d\u5bf9\u4e8e\u8bc6\u522b\u690d\u7269\u75c5\u5bb3\u4efb\u52a1\u6709\u4ec0\u4e48\u7279\u522b\u7684\u597d\u5904\uff1f<\/p>\n\n\n\n\u4f7f\u7528\u201c\u6269\u5f20\u5377\u79ef\u201d\uff08dilated convolution\uff09\u7684\u597d\u5904\u5c31\u662f\u4e3a\u4e86\u6355\u6349\u5168\u5c40\u7279\u5f81\u3002<\/p>\n\n\n\n\u5728\u8bc6\u522b\u690d\u7269\u75c5\u5bb3\u65f6\uff0c\u6a21\u578b\u4e0d\u4ec5\u8981\u770b\u53f6\u7247\u4e0a\u7684\u75c5\u6591\uff08\u5c40\u90e8\u7279\u5f81\uff09\uff0c\u8fd8\u8981\u80fd\u7406\u89e3\u53f6\u7247\u5728\u6574\u4e2a\u590d\u6742\u80cc\u666f\uff08\u5982\u5176\u4ed6\u53f6\u5b50\u3001\u6811\u679d\u3001\u571f\u58e4\uff09\u4e2d\u7684\u60c5\u51b5\uff08\u5168\u5c40\u7279\u5f81\uff09\u3002\u6269\u5f20\u5377\u79ef\u80fd\u5e2e\u52a9\u6a21\u578b\u201c\u770b\u5f97\u66f4\u5e7f\u201d\uff0c\u800c\u4e0d\u4ec5\u4ec5\u662f\u201c\u76ef\u4f4f\u201d\u4e00\u4e2a\u5c0f\u533a\u57df\u3002<\/p>\n\n\n\n\u4e0b\u4e00\u6b65\uff1a\u878d\u5408\u77e5\u8bc6<\/strong><\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nMoreover, we use residual connections to blend new information learned by DCon-Adapter with pretraining prior knowledge.<\/p>\n<\/blockquote>\n\n\n\n\u5728DCon-Adapter\u5904\u7406\u5b8c\u7279\u5f81\u540e\uff0c\u8bba\u6587\u5728 2.4.1 \u8282\u672b\u5c3e\u548c\u56fe 5(A) \u4e2d\u5c55\u793a\u4e86\u4e00\u4e2a\u5173\u952e\u64cd\u4f5c\uff1a\u5b83\u901a\u8fc7\u201c\u6b8b\u5dee\u8fde\u63a5\u201d\uff08residual connections\uff09\u5c06 DCon-Adapter \u5b66\u5230\u7684\u65b0\u7279\u5f81 (fg<\/sub>) \u4e0e CLIP \u539f\u59cb\u7684\u7279\u5f81 (fCLIP<\/sub>) \u7ed3\u5408\u8d77\u6765\u3002\u4e3a\u4ec0\u4e48\u8981\u8fd9\u4e48\u505a\uff1f\u4e3a\u4ec0\u4e48\u4e0d\u53ea<\/em>\u4f7f\u7528 DCon-Adapter \u5b66\u5230\u7684\u65b0\u7279\u5f81\uff0c\u800c\u662f\u8981\u8d39\u529b\u5730\u628a\u5b83\u548c\u539f\u59cb\u7279\u5f81\u52a0\u5728\u4e00\u8d77\u5462\uff1f<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"672\" height=\"410\" src=\"https:\/\/www.leexinghai.com\/aic\/wp-content\/uploads\/2025\/10\/image-30.png\" alt=\"\" class=\"wp-image-3786\" srcset=\"https:\/\/www.leexinghai.com\/aic\/wp-content\/uploads\/2025\/10\/image-30.png 672w, https:\/\/www.leexinghai.com\/aic\/wp-content\/uploads\/2025\/10\/image-30-300x183.png 300w\" sizes=\"auto, (max-width: 672px) 100vw, 672px\" \/><\/figure>\n<\/div>\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u3010\u77e5\u8bc6\u8865\u5145\u3011\u4ec0\u4e48\u662f\u6b8b\u5dee\u8fde\u63a5<\/summary>\n\u6b8b\u5dee\u8fde\u63a5\uff08Residual Connection\uff09\u662f\u4e00\u79cd\u5e38\u7528\u4e8e\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u4e2d\u7684\u7ed3\u6784\uff0c\u7279\u522b\u662f\u5728\u6df1\u5ea6\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\uff08CNN\uff09\u4e2d\u3002\u5b83\u7684\u6838\u5fc3\u601d\u60f3\u662f\u5f15\u5165\u201c\u8df3\u8dc3\u8fde\u63a5\u201d\uff08skip connection\uff09\uff0c\u5373\u5c06\u8f93\u5165\u76f4\u63a5\u4f20\u9012\u5230\u66f4\u6df1\u7684\u7f51\u7edc\u5c42\uff0c\u7ed5\u8fc7\u4e00\u4e9b\u4e2d\u95f4\u5c42\u7684\u8ba1\u7b97\uff0c\u7136\u540e\u4e0e\u8be5\u5c42\u7684\u8f93\u51fa\u76f8\u52a0\u3002\u8fd9\u79cd\u7ed3\u6784\u5e2e\u52a9\u7f51\u7edc\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u51cf\u5c11\u68af\u5ea6\u6d88\u5931\u548c\u68af\u5ea6\u7206\u70b8\u7684\u95ee\u9898\uff0c\u5e76\u4e14\u53ef\u4ee5\u52a0\u901f\u7f51\u7edc\u7684\u6536\u655b\u3002<\/p>\n\n\n\n\u5728\u4f20\u7edf\u7684\u6df1\u5c42\u795e\u7ecf\u7f51\u7edc\u4e2d\uff0c\u968f\u7740\u5c42\u6570\u7684\u589e\u52a0\uff0c\u4fe1\u606f\u5728\u5411\u6df1\u5c42\u4f20\u64ad\u7684\u8fc7\u7a0b\u4e2d\u53ef\u80fd\u4f1a\u9010\u6e10\u4e22\u5931\u6216\u53d8\u5f97\u96be\u4ee5\u4f20\u9012\u3002\u6b8b\u5dee\u8fde\u63a5\u901a\u8fc7\u76f4\u63a5\u5c06\u8f93\u5165\u6dfb\u52a0\u5230\u8f93\u51fa\u4e2d\uff0c\u4fdd\u7559\u4e86\u539f\u59cb\u8f93\u5165\u4fe1\u606f\uff0c\u4f7f\u5f97\u7f51\u7edc\u80fd\u591f\u66f4\u52a0\u6709\u6548\u5730\u8bad\u7ec3\u548c\u4f18\u5316\u3002\u8fd9\u79cd\u65b9\u5f0f\u6700\u65e9\u88ab\u5f15\u5165\u4e8eResNet\uff08Residual Network\uff09\uff0c\u5728ResNet\u4e2d\uff0c\u6b8b\u5dee\u5757\uff08Residual Block\uff09\u5229\u7528\u8fd9\u79cd\u8fde\u63a5\u7ed3\u6784\u6765\u6709\u6548\u5730\u8bad\u7ec3\u975e\u5e38\u6df1\u7684\u7f51\u7edc\u3002<\/p>\n\n\n\n\u7b80\u5355\u6765\u8bf4\uff0c\u6b8b\u5dee\u8fde\u63a5\u7684\u4f5c\u7528\u662f\u901a\u8fc7\u201c\u8df3\u8fc7\u201d\u67d0\u4e9b\u5c42\uff0c\u5141\u8bb8\u4fe1\u606f\u4ece\u7f51\u7edc\u7684\u524d\u9762\u4f20\u9012\u5230\u66f4\u540e\u9762\u7684\u5c42\uff0c\u8fd9\u4f7f\u5f97\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u80fd\u591f\u5b66\u4e60\u5230\u66f4\u6709\u6548\u7684\u7279\u5f81\uff0c\u540c\u65f6\u9632\u6b62\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u51fa\u73b0\u7684\u6027\u80fd\u4e0b\u964d\u95ee\u9898\u3002<\/p>\n<\/details>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nThis approach ensures adaptation to new tasks without losing prior knowledge<\/p>\n\n\n\nTrans:\u8fd9\u79cd\u65b9\u6cd5\u80fd\u786e\u4fdd\u5728\u9002\u5e94\u65b0\u4efb\u52a1\u65f6\u4e0d\u4f1a\u4e22\u5931\u5148\u524d\u7684\u77e5\u8bc6\u3002<\/p>\n<\/blockquote>\n\n\n\n\u8fd9\u5c31\u662f\u201c\u6b8b\u5dee\u8fde\u63a5\u201d (residual connections) \u7684\u6838\u5fc3\u601d\u60f3\uff1a\u786e\u4fdd\u5728\u9002\u5e94\u65b0\u4efb\u52a1\uff08\u5b66\u4e60\u65b0\u77e5\u8bc6 f<\/em>g<\/em><\/sub>\uff09\u65f6\uff0c\u4e0d\u4f1a\u4e22\u5931\u57fa\u7840\u6a21\u578b\u4e2d\u5b9d\u8d35\u7684\u201c\u5148\u9a8c\u77e5\u8bc6\u201d (fCLIP<\/sub><\/em>) \u3002\u8fd9\u79cd\u201c\u65b0\u65e7\u878d\u5408\u201d\u7684\u65b9\u5f0f\uff0c\u8ba9\u6a21\u578b\u65e2\u80fd\u9002\u5e94\u519c\u4e1a\u75c5\u5bb3\u8fd9\u4e2a\u7279\u5b9a\u9886\u57df\uff0c\u53c8\u4fdd\u7559\u4e86 CLIP \u6a21\u578b\u672c\u8eab\u5f3a\u5927\u7684\u6cdb\u5316\u80fd\u529b\u3002\u6211\u4eec\u5df2\u7ecf\u6e05\u695a DCon-Adapter \u662f\u5982\u4f55\u5f3a\u5316\u201c\u56fe\u50cf\u7279\u5f81\u201d\u7684\u4e86\u3002<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nTo guarantee that prior knowledge is not neglected, these new features are combined with the original features fCLIP<\/sub> \u2208 R1* C<\/sup> extracted by the CLIP image encoder via a residual connection. <\/p>\n<\/blockquote>\n\n\n\n\u63a5\u4e0b\u6765\uff0c\u6211\u4eec\u53ef\u4ee5\u770b\u770b\u8bba\u6587\u63d0\u51fa\u7684\u7b2c\u4e8c\u4e2a\u6a21\u5757 WDM\uff08\u6743\u91cd\u5206\u89e3\u77e9\u9635\uff09\uff0c\u5b83\u662f\u5982\u4f55\u4f18\u5316\u201c\u6587\u672c\u7279\u5f81\u201d\u7684 \u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">2.2 WDM\uff08\u6743\u91cd\u5206\u89e3\u77e9\u9635\uff09<\/h2>\n\n\n\n\u6211\u4eec\u5df2\u7ecf\u77e5\u9053 DCon-Adapter \u8d1f\u8d23\u4f18\u5316\u201c\u56fe\u50cf\u7279\u5f81\u201d \ud83d\uddbc\ufe0f\u3002\u800c WDM \u5219\u662f\u7528\u6765\u4f18\u5316\u201c\u6587\u672c\u7279\u5f81\u201d \ud83d\udd21\u3002<\/p>\n\n\n\n\u8bba\u6587\u5728 2.4.2 \u8282 \u4e2d\u63d0\u51fa\u4e86\u4e00\u4e2a\u5f88\u5173\u952e\u7684\u7c7b\u6bd4\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5728\u4f20\u7edf\u7684\u56fe\u50cf\u5206\u7c7b\u4e2d\uff0c\u6a21\u578b\u662f\u5c06\u201c\u56fe\u50cf\u7279\u5f81\u201d\u4e0e\u201c\u5206\u7c7b\u5668\u6743\u91cd\u201d\u76f8\u4e58\u6765\u5f97\u5230\u5206\u6570\u3002<\/li>\n\n\n\n<li>\u800c\u5728 CLIP \u6a21\u578b\u4e2d\uff0c\u6a21\u578b\u662f\u5c06\u201c\u56fe\u50cf\u7279\u5f81\u201d\u4e0e\u201c\u6587\u672c\u7279\u5f81\u201d\uff08\u6216\u79f0\u4e3a\u201c\u63d0\u793a\u8bcd\u5d4c\u5165\u201d\uff09\u8fdb\u884c\u6bd4\u8f83\u6765\u5f97\u5230\u5206\u6570\u3002<\/li>\n<\/ul>\n\n\n\n\u6839\u636e\u8fd9\u4e2a\u7c7b\u6bd4\uff0c\u8bba\u6587\u8ba4\u4e3a CLIP \u4e2d\u7684\u201c\u6587\u672c\u7279\u5f81\u201d\u5728\u529f\u80fd\u4e0a\u626e\u6f14\u4e86\u4ec0\u4e48\u89d2\u8272\uff1f<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nIn deep neural networks, image classification is typically achieved by multiplying the image features with the classifier weights, resulting in a score matrix. This matrix is then transformed into a probability matrix via the SoftMax function [58], with the class label being determined by the index corresponding to the maximum value in the matrix. In CLIP, a similarity matrix is computed between image features and text features for each class. The class label is determined by the text with the highest similarity to the image. A comparison reveals that the embeddings extracted by the text encoder function similarly to those extracted by the classifier in image classification. Therefore, the prompt embeddings can be understood as the weights of the classifier.<\/p>\n\n\n\nTrans:\u5728\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u4e2d\uff0c\u56fe\u50cf\u5206\u7c7b\u901a\u5e38\u901a\u8fc7\u5c06\u56fe\u50cf\u7279\u5f81\u4e0e\u5206\u7c7b\u5668\u6743\u91cd\u76f8\u4e58\u6765\u5b9e\u73b0\uff0c\u4ece\u800c\u751f\u6210\u4e00\u4e2a\u5206\u6570\u77e9\u9635\u3002\u8be5\u77e9\u9635\u968f\u540e\u901a\u8fc7SoftMax\u51fd\u6570[58]\u8f6c\u6362\u4e3a\u6982\u7387\u77e9\u9635\uff0c\u7c7b\u522b\u6807\u7b7e\u7531\u77e9\u9635\u4e2d\u6700\u5927\u503c\u5bf9\u5e94\u7684\u7d22\u5f15\u51b3\u5b9a\u3002\u5728CLIP\u6a21\u578b\u4e2d\uff0c\u7cfb\u7edf\u4f1a\u4e3a\u6bcf\u4e2a\u7c7b\u522b\u8ba1\u7b97\u56fe\u50cf\u7279\u5f81\u4e0e\u6587\u672c\u7279\u5f81\u4e4b\u95f4\u7684\u76f8\u4f3c\u5ea6\u77e9\u9635\u3002\u7c7b\u522b\u6807\u7b7e\u7531\u4e0e\u56fe\u50cf\u76f8\u4f3c\u5ea6\u6700\u9ad8\u7684\u6587\u672c\u786e\u5b9a\u3002\u5bf9\u6bd4\u53d1\u73b0\uff0c\u6587\u672c\u7f16\u7801\u5668\u63d0\u53d6\u7684\u5d4c\u5165\u5411\u91cf\u4e0e\u5206\u7c7b\u5668\u5728\u56fe\u50cf\u5206\u7c7b\u4e2d\u63d0\u53d6\u7684\u5d4c\u5165\u5411\u91cf\u5177\u6709\u76f8\u4f3c\u6027\u3002\u56e0\u6b64\uff0c\u63d0\u793a\u8bcd\u5d4c\u5165\u53ef\u7406\u89e3\u4e3a\u5206\u7c7b\u5668\u7684\u6743\u91cd\u53c2\u6570\u3002<\/p>\n<\/blockquote>\n\n\n\n\u8bba\u6587\u5c31\u662f\u8fd9\u4e48\u770b\u7684\uff1aCLIP \u91cc\u7684\u201c\u63d0\u793a\u8bcd\u5d4c\u5165\u201d\uff08text embeddings\uff09\u529f\u80fd\u4e0a\u5c31\u50cf\u662f\u5206\u7c7b\u5668\u7684\u201c\u6743\u91cd\u201d \u3002<\/p>\n\n\n\n\u90a3\u4e48\uff0c\u65e2\u7136\u5728\u5c11\u6837\u672c\u4efb\u52a1\u4e2d\uff0c\u6211\u4eec\u4e0d\u60f3\uff08\u4e5f\u4e0d\u80fd\uff09\u53bb\u5fae\u8c03\u6574\u4e2a\u5e9e\u5927\u7684\u6a21\u578b\uff0c\u800c\u4e14\u6211\u4eec\u53c8\u60f3\u8c03\u6574\u8fd9\u4e9b\u201c\u6743\u91cd\u201d\u8ba9\u5b83\u4eec\u66f4\u9002\u5e94\u6211\u4eec\u7684\u4efb\u52a1\uff0c\u8fd9\u5c31\u4f1a\u5bfc\u81f4\u4e00\u4e2a\u4ec0\u4e48\u95ee\u9898\u5462\uff1f\uff08\u63d0\u793a\uff1a\u60f3\u60f3\u201c\u53c2\u6570\u201d\u548c\u201c\u8fc7\u62df\u5408\u201d\u7684\u5173\u7cfb\uff09<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nImportantly, excessive parameters can lead to overfitting in few-shot tasks.<\/p>\n\n\n\nTrans:\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u53c2\u6570\u8fc7\u591a\u5bb9\u6613\u5bfc\u81f4\u5c11\u6837\u672c\u4efb\u52a1\u51fa\u73b0\u8fc7\u62df\u5408\u73b0\u8c61\u3002<\/p>\n<\/blockquote>\n\n\n\n\u5982\u679c\u6211\u4eec\u76f4\u63a5\u5fae\u8c03 CLIP \u7684\u201c\u6587\u672c\u7279\u5f81\u201d\uff08\u63d0\u793a\u8bcd\u5d4c\u5165\uff09\uff0c\u5c31\u8981\u8c03\u6574\u5927\u91cf\u7684\u53c2\u6570\u3002\u800c\u5728\u201c\u5c11\u6837\u672c\u201d\u4efb\u52a1\u4e2d\uff0c\u53ef\u8bad\u7ec3\u7684\u53c2\u6570\u592a\u591a\uff0c\u6837\u672c\u592a\u5c11\uff0c\u5c31\u6781\u6613\u5bfc\u81f4\u8fc7\u62df\u5408\u3002\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\uff0c\u8bba\u6587\u5c31\u5f15\u5165\u4e86 WDM\uff08\u6743\u91cd\u5206\u89e3\u77e9\u9635\uff09\u3002\u5b83\u91c7\u7528\u4e86\u4e00\u79cd\u5de7\u5999\u7684\u201c\u964d\u7ef4\u201d\u601d\u60f3 \uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u5b83\u4e0d\u53bb\u76f4\u63a5\u5b66\u4e60\u4e00\u4e2a\u5de8\u5927\u7684\u3001\u5b8c\u6574\u7684\u53ef\u8bad\u7ec3\u77e9\u9635 M\u3002<\/li>\n\n\n\n<li>\u800c\u662f\u5b66\u4e60\u4e24\u4e2a\u66f4\u5c0f\u7684\u201c\u4f4e\u79e9\u77e9\u9635\u201d A \u548c B\uff0c\u4f7f\u5f97 M = A * B\u3002<\/li>\n<\/ol>\n\n\n\n\u901a\u8fc7\u8fd9\u79cd\u65b9\u5f0f\uff0c\u9700\u8981\u5b66\u4e60\u7684\u53c2\u6570\u603b\u91cf\u88ab\u5927\u5927\u51cf\u5c11\u4e86\u3002<\/p>\n\n\n\n\u6211\u4eec\u5df2\u7ecf\u4e86\u89e3\u4e86\u4e24\u4e2a\u6838\u5fc3\u6a21\u5757\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>DCon-Adapter\uff1a\u901a\u8fc7\u6b8b\u5dee\u8fde\u63a5\uff0c\u5728\u4e0d\u4e22\u5931\u5148\u9a8c\u77e5\u8bc6\u7684\u524d\u63d0\u4e0b\uff0c\u5f3a\u5316\u201c\u56fe\u50cf\u7279\u5f81\u201d\u3002<\/li>\n\n\n\n<li>WDM\uff1a\u901a\u8fc7\u4f4e\u79e9\u5206\u89e3\uff0c\u7528\u5f88\u5c11\u7684\u53c2\u6570\u9ad8\u6548\u5730\u5fae\u8c03\u201c\u6587\u672c\u7279\u5f81\u201d\u3002<\/li>\n<\/ul>\n\n\n\n\u8fd9\u4e24\u4e2a\u6a21\u5757\u534f\u540c\u5de5\u4f5c\uff0c\u4f7f\u5f97 PlantCaFo \u80fd\u591f\u5728\u4e0d\u5fae\u8c03\u6574\u4e2a\u5927\u6a21\u578b\u7684\u60c5\u51b5\u4e0b\uff0c\u9ad8\u6548\u5730\u9002\u5e94\u690d\u7269\u75c5\u5bb3\u8bc6\u522b\u4efb\u52a1\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity is-style-wide\"\/>\n\n\n\n<h2 class=\"wp-block-heading has-text-align-center\">3.\u5b9e\u9a8c\u7ed3\u679c<\/h2>\n\n\n\n\u6211\u4eec\u6765\u770b\u770b\u5b9e\u9a8c\u7ed3\u679c \ud83d\udcca\u3002<\/p>\n\n\n\n\u8bba\u6587\u5728\u51e0\u4e2a\u6570\u636e\u96c6\u4e0a\u6d4b\u8bd5\u4e86 PlantCaFo\uff0c\u4e3b\u8981\u662f\u5728 \"N-way K-shot\" \u8bbe\u7f6e\u4e0b\u6bd4\u8f83\u51c6\u786e\u7387\u3002<\/p>\n\n\n\n\u6211\u4eec\u53ef\u4ee5\u4ece\u51e0\u4e2a\u65b9\u9762\u6765\u770b\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u4e3b\u8981\u6027\u80fd\u5bf9\u6bd4\uff1a\u5728 PlantVillage \u548c Cassava \u6570\u636e\u96c6\u4e0a\uff0cPlantCaFo \u548c\u5176\u4ed6\u65b9\u6cd5\uff08\u5982 CaFo-Base, Tip-Adapter-F\uff09\u7684\u51c6\u786e\u7387\u5bf9\u6bd4\uff08\u56fe7\uff0c\u88682\u548c\u88683\uff09\u3002<\/li>\n\n\n\n<li>\u6a21\u5757\u8d21\u732e\u5206\u6790\uff08\u6d88\u878d\u5b9e\u9a8c\uff09\uff1aDCon-Adapter \u548c WDM \u8fd9\u4e24\u4e2a\u6a21\u5757\u5230\u5e95\u5404\u81ea\u63d0\u5347\u4e86\u591a\u5c11\u6027\u80fd\uff08\u88687\uff09\uff1f<\/li>\n\n\n\n<li>\u6cdb\u5316\u80fd\u529b\uff1a\u6a21\u578b\u5728\u201c\u5206\u5e03\u5916\u201d\u6570\u636e\u96c6\uff08PDL\uff09\u4e0a\u7684\u8868\u73b0\u5982\u4f55\uff08\u88686\uff09\uff1f<\/li>\n<\/ol>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">3.1\u4e3b\u8981\u6027\u80fd\u5bf9\u6bd4<\/h2>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"717\" height=\"314\" src=\"https:\/\/www.leexinghai.com\/aic\/wp-content\/uploads\/2025\/10\/image-31.png\" alt=\"\" class=\"wp-image-3799\" srcset=\"https:\/\/www.leexinghai.com\/aic\/wp-content\/uploads\/2025\/10\/image-31.png 717w, https:\/\/www.leexinghai.com\/aic\/wp-content\/uploads\/2025\/10\/image-31-300x131.png 300w\" sizes=\"auto, (max-width: 717px) 100vw, 717px\" \/><\/figure>\n<\/div>\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"363\" height=\"294\" src=\"https:\/\/www.leexinghai.com\/aic\/wp-content\/uploads\/2025\/10\/image-32.png\" alt=\"\" class=\"wp-image-3800\" srcset=\"https:\/\/www.leexinghai.com\/aic\/wp-content\/uploads\/2025\/10\/image-32.png 363w, https:\/\/www.leexinghai.com\/aic\/wp-content\/uploads\/2025\/10\/image-32-300x243.png 300w\" sizes=\"auto, (max-width: 363px) 100vw, 363px\" \/><\/figure>\n<\/div>\n\n\n\"PlantCaFo*\" (\u5e26\u661f\u53f7\u7684\u7248\u672c) \u662f\u5728 PlantCaFo \u7684\u57fa\u7840\u4e0a\uff0c\u989d\u5916\u4f7f\u7528\u4e86\u4e24\u79cd\u6570\u636e\u589e\u5f3a\uff08data augmentation\uff09\u6280\u672f\u8fdb\u884c\u8bad\u7ec3\u7684\u7248\u672c\uff0c\u5177\u4f53\u6765\u8bf4\u5c31\u662f Mixup \u548c CutMix \u3002<\/p>\n\n\n\n\u5728\u88682\u4e2d\uff0cPlantCaFo* \u7684\u51c6\u786e\u7387 (94.23%) \u751a\u81f3\u6bd4 PlantCaFo (93.53%) \u8fd8\u8981\u9ad8 \u3002<\/p>\n\n\n\n\u8bba\u6587\u5728 2.5 \u8282\uff08\u56fe6\uff09\u4e13\u95e8\u89e3\u91ca\u4e86\u8fd9\u4e24\u79cd\u6280\u672f\u3002<\/p>\n\n\n\nMixup \u548c CutMix \u90fd\u662f\u6570\u636e\u589e\u5f3a\u6280\u672f\uff0c\u7528\u6765\u201c\u51ed\u7a7a\u201d\u521b\u9020\u66f4\u591a\u7684\u8bad\u7ec3\u6837\u672c\u3002<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nMixup is a data augmentation technique based on linear interpola\u0002tion. It generates new samples by linearly interpolating two different training examples in a batch, along with their labels<\/p>\n\n\n\nCutMix involves cropping out sections of an image and randomly filling them with regions from other images in the training set. Labels are allocated proportionally. Mixup uses information from the entire image to merge two images, whereas CutMix mixes images by cropping and pasting parts of the image.<\/p>\n<\/blockquote>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Mixup\uff1a<\/strong>\u5b83\u901a\u8fc7\u7ebf\u6027\u63d2\u503c\uff08linear interpolation\uff09\u6765\u6df7\u5408\u4e24\u5f20\u4e0d\u540c\u7684\u56fe\u50cf\u53ca\u5176\u6807\u7b7e \u3002\u7b80\u5355\u8bf4\uff0c\u5c31\u662f\u628a\u4e24\u5f20\u56fe\u7247\u201c\u53e0\u201d\u5728\u4e00\u8d77\uff0c\u4e00\u5f20\u5360 70% \u900f\u660e\u5ea6\uff0c\u53e6\u4e00\u5f20\u5360 30%\uff08\u8fd9\u4e2a\u6bd4\u4f8b\u662f\u968f\u673a\u7684 \uff09\uff0c\u6807\u7b7e\u4e5f\u6309\u8fd9\u4e2a\u6bd4\u4f8b\u6df7\u5408\u3002<\/li>\n\n\n\n<li>CutMix\uff1a<\/strong>\u5b83\u4f1a\u4ece\u4e00\u5f20\u56fe\u50cf\u4e2d\u201c\u526a\u5207\u201d\u51fa\u4e00\u5757\u533a\u57df\uff0c\u7136\u540e\u201c\u7c98\u8d34\u201d\u5230\u53e6\u4e00\u5f20\u56fe\u50cf\u4e0a\uff0c\u6807\u7b7e\u4e5f\u6309\u526a\u5207\u7c98\u8d34\u7684\u9762\u79ef\u6bd4\u4f8b\u6765\u5206\u914d \u3002<\/li>\n<\/ul>\n\n\n\n\u8bba\u6587\u5728 2.5 \u8282\u4e2d\u5206\u522b\u89e3\u91ca\u4e86\u8fd9\u4e24\u79cd\u65b9\u6cd5\u4e3a\u4ec0\u4e48\u6709\u76ca\u3002\u4f8b\u5982\uff0cMixup \u80fd\u201c\u5f15\u5165\u566a\u58f0\u548c\u5e72\u6270\uff0c\u4ee5\u589e\u52a0\u6a21\u578b\u7684\u9c81\u68d2\u6027\u201d \u3002<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nMixup but uses a parameter to control the cropping size. CutMix requires the model to recognize objects from a local view and adds information from other samples into the cropped region, which can enhance the localization ability of the model and improve its classification performance.<\/p>\n\n\n\nTrans:Mixup\u867d\u91c7\u7528\u53c2\u6570\u63a7\u5236\u88c1\u526a\u5c3a\u5bf8\uff0c\u4f46CutMix\u6a21\u578b\u9700\u5148\u8bc6\u522b\u5c40\u90e8\u89c6\u56fe\u4e2d\u7684\u7269\u4f53\uff0c\u5e76\u5c06\u5176\u4ed6\u6837\u672c\u4fe1\u606f\u878d\u5165\u88c1\u526a\u533a\u57df\uff0c\u8fd9\u79cd\u673a\u5236\u80fd\u589e\u5f3a\u6a21\u578b\u7684\u5b9a\u4f4d\u80fd\u529b\uff0c\u4ece\u800c\u63d0\u5347\u5206\u7c7b\u6027\u80fd\u3002<\/p>\n<\/blockquote>\n\n\n\nCutMix \u901a\u8fc7\u8fd9\u79cd\u201c\u526a\u5207\u7c98\u8d34\u201d\u7684\u65b9\u5f0f\uff0c\u8feb\u4f7f\u6a21\u578b\u4ece\u4e00\u4e2a\u5c40\u90e8\u89c6\u56fe\uff08a local view\uff09\u4e2d\u8bc6\u522b\u7269\u4f53\uff0c\u5e76\u4e14\u5728\u526a\u5207\u533a\u57df\u52a0\u5165\u4e86\u5176\u4ed6\u6837\u672c\u7684\u4fe1\u606f \u3002\u8fd9\u5c31\u80fd\u589e\u5f3a\u6a21\u578b\u7684\u5b9a\u4f4d\u80fd\u529b\uff08localization ability\uff09\uff0c\u4ece\u800c\u63d0\u5347\u5206\u7c7b\u6027\u80fd \u3002<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nThe incorporation of Mixup and CutMix augmentations further boosts the model's performance by enhancing its ability to generalize across different plant disease types, which likely contributes to the increased performance in the PlantCaFo* variant.<\/p>\n\n\n\n\u6df7\u5408\u589e\u5f3a\u548c\u526a\u5207\u589e\u5f3a\u7684\u6574\u5408\u901a\u8fc7\u589e\u5f3a\u6a21\u578b\u5728\u4e0d\u540c\u690d\u7269\u75c5\u5bb3\u7c7b\u578b\u4e0a\u7684\u6cdb\u5316\u80fd\u529b\uff0c\u8fdb\u4e00\u6b65\u63d0\u5347\u4e86\u5176\u6027\u80fd\u8868\u73b0\uff0c\u8fd9\u53ef\u80fd\u662fPlantCaFo*\u53d8\u4f53\u6027\u80fd\u63d0\u5347\u7684\u5173\u952e\u56e0\u7d20\u3002<\/p>\n<\/blockquote>\n\n\n\n\u6240\u4ee5\uff0c\"PlantCaFo*\" (\u5e26\u661f\u53f7\u7684\u7248\u672c) \u5c31\u662f\u901a\u8fc7\u7efc\u5408\u8fd0\u7528 Mixup \u548c CutMix \u8fd9\u4e24\u79cd\u6570\u636e\u589e\u5f3a\u65b9\u6cd5\uff0c\u8fdb\u4e00\u6b65\u63d0\u5347\u4e86\u6a21\u578b\u7684\u9c81\u68d2\u6027\u548c\u6027\u80fd \u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">3.2 \u6a21\u5757\u8d21\u732e\u5206\u6790\uff08\u6d88\u878d\u5b9e\u9a8c\uff09<\/h2>\n\n\n\n\u8868 7 (Table 7)<\/strong>\u5c31\u662f\u201c\u6d88\u878d\u5b9e\u9a8c\u201d\uff08Ablation studies\uff09\u3002<\/p>\n\n\n\n\u8fd9\u79cd\u5b9e\u9a8c\u5c31\u50cf\u662f\u201c\u642d\u79ef\u6728\u201d\u548c\u201c\u62c6\u79ef\u6728\u201d\uff0c\u901a\u8fc7\u201c\u6dfb\u52a0\u201d\u6216\u201c\u79fb\u9664\u201d\u6a21\u578b\u7684\u67d0\u4e2a\u7ec4\u4ef6\uff0c\u6765\u770b\u770b\u5b83\u5bf9\u6700\u7ec8\u6027\u80fd\uff08\u51c6\u786e\u7387\uff09\u6709\u591a\u5927\u8d21\u732e\u3002<\/p>\n\n\n\n\u8ba9\u6211\u4eec\u6765\u505a\u4e00\u4e2a\u5bf9\u6bd4\u3002\u8bf7\u770b\u8868 7\uff1a<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"703\" height=\"155\" src=\"https:\/\/www.leexinghai.com\/aic\/wp-content\/uploads\/2025\/10\/image-33.png\" alt=\"\" class=\"wp-image-3801\" srcset=\"https:\/\/www.leexinghai.com\/aic\/wp-content\/uploads\/2025\/10\/image-33.png 703w, https:\/\/www.leexinghai.com\/aic\/wp-content\/uploads\/2025\/10\/image-33-300x66.png 300w\" sizes=\"auto, (max-width: 703px) 100vw, 703px\" \/><\/figure>\n<\/div>\n\n\n<ul class=\"wp-block-list\">\n<li>\u7b2c 1 \u884c\uff1a\u662f\u57fa\u7ebf\u6a21\u578b\uff08CaFo-Base\uff09\u3002<\/li>\n\n\n\n<li>\u7b2c 3 \u884c\uff1a\u662f\u57fa\u7ebf\u6a21\u578b + DCon-Adapter\u3002<\/li>\n\n\n\n<li>\u7b2c 4 \u884c\uff1a\u662f\u57fa\u7ebf\u6a21\u578b + WDM\u3002<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>DCon-Adapter\uff08\u4e3a\u56fe\u50cf\u7279\u5f81\u6dfb\u52a0\u65b0\u53c2\u6570\uff09\u5728 1-shot \u548c 2-shot \u65f6\u8868\u73b0\u4e0d\u4f73\uff0c\u4f46\u968f\u7740\u6837\u672c\u589e\u52a0\uff084, 8, 16-shot\uff09\uff0c\u5b83\u7684\u6027\u80fd\u63d0\u5347\u975e\u5e38\u660e\u663e \u3002<\/li>\n\n\n\n<li>WDM\uff08\u4e3a\u6587\u672c\u7279\u5f81\u6dfb\u52a0\u65b0\u53c2\u6570\uff09\u5728 1-shot \u548c 2-shot \u65f6\u8868\u73b0\u76f8\u5bf9\u66f4\u597d \u3002<\/li>\n<\/ul>\n\n\n\n\u8bba\u6587\u5728 3.3.1 \u8282\u4e2d\u4e5f\u63d0\u5230\u4e86\u8fd9\u4e00\u70b9\uff0c\u5373\u5f53\u6837\u672c\u6570\u91cf\u975e\u5e38\u5c11\uff08\u4f8b\u5982 1 \u6216 2 \u4e2a\u6837\u672c\uff09\u65f6\uff0c\u8fd9\u4e9b\u5e26\u6709\u53ef\u8bad\u7ec3\u53c2\u6570\u7684\u6a21\u5757\u7684\u80fd\u529b\u4f1a\u53d7\u5230\u9650\u5236 \u3002<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nhowever, this combination may be limited when the number of samples is small (e.g., 1 or 2 samples) because of the learning ability of the trainable parameters.<\/p>\n<\/blockquote>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8003\u8651\u5230 DCon-Adapter \u662f\u4e00\u4e2a\u9700\u8981 \u5b66\u4e60 \u65b0\u77e5\u8bc6\u7684\u6a21\u5757\uff0c\u4e3a\u4ec0\u4e48\u5f53\u5b83\u53ea\u6709 1 \u6216 2 \u4e2a\u6837\u672c\u65f6\uff0c\u8868\u73b0\u4f1a\u53cd\u800c\u4e0d\u5982 WDM \u5462\uff1f\n<ul class=\"wp-block-list\">\n<li>\u56e0\u4e3a\u6837\u672c\u592a\u5c11\u6240\u4ee5\u5b66\u4e60\u7684\u80fd\u529b\u6709\u9650<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n\u4e00\u4e2a\u6a21\u5757\uff08\u5982 DCon-Adapter\uff09\u88ab\u8bbe\u8ba1\u7528\u6765\u5b66\u4e60\u65b0\u7684\u89c6\u89c9\u7279\u5f81\uff0c\u4f46\u5f53\u5b83\u53ea\u6709 1 \u6216 2 \u4e2a\u6837\u672c\u65f6\uff0c\u5b83\u6ca1\u6709\u8db3\u591f\u7684\u4fe1\u606f\u6765\u6709\u6548\u5b66\u4e60\u3002\u8fd9\u5c31\u50cf\u662f\u8ba9\u4f60\u53ea\u770b\u4e00\u5f20\u7167\u7247\u5c31\u5b66\u4f1a\u8bc6\u522b\u4e00\u79cd\u4f60\u4ece\u672a\u89c1\u8fc7\u7684\u590d\u6742\u75c5\u5bb3\u3002<\/p>\n\n\n\n\u8bba\u6587\u4e5f\u63d0\u5230\u4e86\u8fd9\u4e2a\u5c40\u9650\uff1a\u201cthis combination may be limited when the number of samples is small (e.g., 1 or 2 samples) because of the learning ability of the trainable parameters\u201d \u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4e0d\u8fc7\uff0c\u6211\u4eec\u4e4b\u524d\u8ba8\u8bba\u8fc7\u4e00\u4e2a\u89e3\u51b3\u65b9\u6848\u3002\u8bf7\u518d\u770b\u4e00\u4e0b \u8868 7 \u7684\u6700\u540e\u4e00\u884c\uff08AG \u5217\uff09\uff0c\u8bba\u6587\u5f15\u5165\u4e86\u4ec0\u4e48\u65b9\u6cd5\u6765\u89e3\u51b3\u4e86\u8fd9\u4e2a\u95ee\u9898\uff0c\u4f7f\u5f97\u6a21\u578b\u5728 1-shot \u548c 2-shot \u4e0a\u7684\u6027\u80fd\u4e5f\u5f97\u5230\u4e86\u63d0\u5347\uff1f\n<ul class=\"wp-block-list\">\n<li>\u5c31\u662f AG\uff08Data Augmentation\uff0c\u6570\u636e\u589e\u5f3a\uff09\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n\u8fd8\u8bb0\u5f97\u6211\u4eec\u4e4b\u524d\u8ba8\u8bba\u8fc7\u7684 Mixup \u548c CutMix \u5417\uff1f\u5b83\u4eec\u901a\u8fc7\u201c\u6df7\u5408\u201d\u548c\u201c\u526a\u5207\u201d\u56fe\u50cf\u6765\u521b\u9020\u65b0\u7684\u8bad\u7ec3\u6837\u672c\u3002<\/p>\n\n\n\n\u5f53 DCon-Adapter \u53ea\u6709 1 \u6216 2 \u4e2a\u771f\u5b9e\u6837\u672c\u65f6\uff0c\u5b83\u5f88\u96be\u5b66\u4e60 \uff1b\u4f46\u6570\u636e\u589e\u5f3a\u6280\u672f\uff08AG\uff09\u7ed9\u5b83\u63d0\u4f9b\u4e86\u66f4\u591a\u201c\u4eba\u9020\u201d\u7684\u6837\u672c\u8fdb\u884c\u7ec3\u4e60\uff0c\u8fd9\u6709\u6548\u5730\u89e3\u51b3\u4e86\u6837\u672c\u592a\u5c11\u3001\u5b66\u4e60\u80fd\u529b\u6709\u9650\u7684\u95ee\u9898 \u3002<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nFinally, this issue has been effectively addressed by introducing data augmentation techniques.<\/p>\n<\/blockquote>\n\n\n\n\u6211\u4eec\u5df2\u7ecf\u5206\u6790\u4e86\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u4e3b\u8981\u6027\u80fd\uff1aPlantCaFo \u4f18\u4e8e\u57fa\u7ebf\u6a21\u578b\u3002<\/li>\n\n\n\n<li>\u6d88\u878d\u5b9e\u9a8c\uff1aDCon-Adapter \u548c WDM \u786e\u5b9e\u6709\u6548\uff0c\u5e76\u4e14\u6570\u636e\u589e\u5f3a\uff08AG\uff09\u89e3\u51b3\u4e86\u5c0f\u6837\u672c\u4e0b\u7684\u5b66\u4e60\u9650\u5236\u3002<\/li>\n<\/ol>\n\n\n\n\u5b9e\u9a8c\u90e8\u5206\u8fd8\u5269\u4e0b\u4e00\u4e2a\u5173\u952e\u95ee\u9898\uff1a\u6cdb\u5316\u80fd\u529b\u3002<\/p>\n\n\n\n\u6a21\u578b\u5728\u8bad\u7ec3\u96c6\uff08PlantVillage\uff09\u4e0a\u8868\u73b0\u597d\u662f\u4e00\u56de\u4e8b\uff0c\u4f46\u5982\u679c\u628a\u5b83\u7528\u5728\u4e00\u4e2a\u5b83\u4ece\u672a\u89c1\u8fc7\u7684\u65b0\u6570\u636e\u96c6\uff08\u201c\u5206\u5e03\u5916\u201d\u6570\u636e\u96c6 PDL\uff09\u4e0a\uff0c\u5b83\u8fd8\u80fd\u7528\u5417\uff1f<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">3.3 \u6cdb\u5316\u80fd\u529b<\/h2>\n\n\n\n\u6211\u4eec\u6765\u770b\u8868 6 (Table 6)<\/strong>\uff0c\u8fd9\u662f\u5173\u4e8e\u201c\u6cdb\u5316\u80fd\u529b\u201d\uff08Generalization ability\uff09\u7684\u5b9e\u9a8c\u3002<\/p>\n\n\n\n\u8fd9\u4e2a\u5b9e\u9a8c\u8bbe\u7f6e\u662f\u8fd9\u6837\u7684\uff1a<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nThe models are trained on PlantVillage (source domain) with \u201c8-way 4-shot\u201d, \u201c8-way 8-shot\u201d, \u201c8-way 16-shot\u201d, \u201c13-way 4-shot\u201d, \u201c13-way 8-shot\u201d and \u201c13-way 16-shot\u201d settings and then tested on split1 and split2 of PDL.<\/p>\n<\/blockquote>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bad\u7ec3\uff08\u6e90\u57df\uff09\uff1a<\/strong>\u6a21\u578b\u5728 PlantVillage \u6570\u636e\u96c6\u4e0a\u8bad\u7ec3 \u3002<\/li>\n\n\n\n<li>\u6d4b\u8bd5\uff08\u76ee\u6807\u57df\uff09\uff1a<\/strong>\u7136\u540e\u5728 PDL \u6570\u636e\u96c6\u4e0a\u6d4b\u8bd5\u3002PDL \u662f\u4e00\u4e2a\u201c\u5206\u5e03\u5916\u201d\uff08out-of-distribution\uff09\u6570\u636e\u96c6\uff0c\u610f\u5473\u7740\u5b83\u7684\u6570\u636e\u5206\u5e03\u548c PlantVillage \u4e0d\u4e00\u6837\uff08\u6bd4\u5982\u80cc\u666f\u66f4\u590d\u6742\u3001\u62cd\u6444\u573a\u666f\u4e0d\u540c\uff09\u3002<\/li>\n<\/ul>\n\n\n\nPDL \u6570\u636e\u96c6\u88ab\u5206\u6210\u4e86\u4e24\u90e8\u5206 \uff1a<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nTo evaluate the generalization ability of our model, we conduct ex-periments using an out-of distribution dataset (PDL). We divide PDL into split1 and split2, as shown in Table 5. Split1 consists of multiple diseases from a single plant species, whereas split2 includes multiple diseases from various plant species. <\/p>\n<\/blockquote>\n\n\n\nsplit1\uff1a\u53ea\u5305\u542b\u591a\u79cd\u756a\u8304\u7684\u75c5\u5bb3\uff08\u540c\u4e00\u7269\u79cd\uff0c\u4e0d\u540c\u75c5\u5bb3\uff09\u3002<\/p>\n\n\n\nsplit2\uff1a\u5305\u542b\u82f9\u679c\u3001\u7389\u7c73\u3001\u8461\u8404\u7b49\u591a\u79cd\u4f5c\u7269\u7684\u75c5\u5bb3\uff08\u4e0d\u540c\u7269\u79cd\uff0c\u4e0d\u540c\u75c5\u5bb3\uff09\u3002<\/p>\n\n\n\n\u73b0\u5728\u8bf7\u770b\u8868 6\uff0c\u5728 split1\uff08\u756a\u8304\uff09\u4e0a\uff0c\u6211\u4eec\u7684 PlantCaFo \u548c PlantCaFo* \u4e0e\u57fa\u7ebf\u6a21\u578b (CaFo-Base) \u76f8\u6bd4\uff0c\u8868\u73b0\u5982\u4f55\uff1f<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full is-resized\"><img loading=\"lazy\" decoding=\"async\" width=\"345\" height=\"134\" src=\"https:\/\/www.leexinghai.com\/aic\/wp-content\/uploads\/2025\/10\/image-34.png\" alt=\"\" class=\"wp-image-3805\" style=\"width:565px;height:auto\" srcset=\"https:\/\/www.leexinghai.com\/aic\/wp-content\/uploads\/2025\/10\/image-34.png 345w, https:\/\/www.leexinghai.com\/aic\/wp-content\/uploads\/2025\/10\/image-34-300x117.png 300w\" sizes=\"auto, (max-width: 345px) 100vw, 345px\" \/><\/figure>\n<\/div>\n\n\n\u5728 split1\uff08\u756a\u8304\u6570\u636e\u96c6\uff09\u4e0a\uff0cPlantCaFo \u548c PlantCaFo* \u76f8\u6bd4\u57fa\u7ebf\u6a21\u578b (CaFo-Base) \u6709\u975e\u5e38\u663e\u8457\u7684\u63d0\u5347\u3002<\/p>\n\n\n\n\u8fd9\u8868\u660e\uff0c\u5f53\u6a21\u578b\u9700\u8981\u6cdb\u5316\u5230\u201c\u540c\u4e00\u79cd\u7c7b\u3001\u4e0d\u540c\u75c5\u5bb3\u201d\u7684\u4efb\u52a1\u65f6\uff08\u90fd\u662f\u756a\u8304\uff09\uff0c\u6211\u4eec\u7684\u65b0\u6a21\u5757\uff08DCon-Adapter \u548c WDM\uff09\u975e\u5e38\u6709\u6548\u3002<\/p>\n\n\n\n\u73b0\u5728\uff0c\u6211\u4eec\u6765\u770b\u770b\u66f4\u6709\u6311\u6218\u6027\u7684 split2\uff08\u5305\u542b\u82f9\u679c\u3001\u7389\u7c73\u3001\u8461\u8404\u7b49\uff09\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5728 split2 \u4e0a\uff0cPlantCaFo \u548c PlantCaFo* \u7684\u8868\u73b0\u4e0e\u57fa\u7ebf (CaFo-Base) \u76f8\u6bd4\u53c8\u5982\u4f55\u5462\uff1f\n<ul class=\"wp-block-list\">\n<li>\u9664\u4e8616shot\u6bd4\u57fa\u7ebf\u6a21\u578b\u597d\uff0c4\u548c8shot\u90fd\u6bd4\u57fa\u7ebf\u6a21\u578b\u5dee<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n\u8fd9\u6b63\u662f\u4e00\u4e2a\u975e\u5e38\u5173\u952e\u4e14\u53cd\u76f4\u89c9\u7684\u53d1\u73b0\u3002\u5728 split2\uff08\u591a\u7269\u79cd\uff09\u4e0a\uff0c\u6211\u4eec\u7684\u6a21\u578b\u5728 4-shot \u548c 8-shot \u8bbe\u7f6e\u4e0b\uff0c\u8868\u73b0\u4e0d\u5982\u57fa\u7ebf\u6a21\u578b CaFo-Base \u3002<\/p>\n\n\n\n\u8bba\u6587\u5728 3.2.3 \u8282\u4e2d\u4e13\u95e8\u8ba8\u8bba\u4e86\u8fd9\u4e2a\u73b0\u8c61\u3002\u4f5c\u8005\u5c06\u5176\u5f52\u56e0\u4e8e\u8bad\u7ec3\u6570\u636e\uff08PlantVillage\uff09\u548c split2 \u6d4b\u8bd5\u6570\u636e\u4e4b\u95f4\u7684\u5de8\u5927\u5dee\u5f02\u3002<\/p>\n\n\n\n\u6839\u636e\u8bba\u6587\u7684\u5206\u6790\uff0csplit2 \u7684\u6570\u636e\u4e0e\u7528\u4e8e\u8bad\u7ec3\u7684 PlantVillage \u6570\u636e\u96c6\u76f8\u6bd4\uff0c\u6709\u4ec0\u4e48\u5173\u952e\u7684\u5dee\u5f02\uff0c\u5bfc\u81f4\u4e86\u8fd9\u79cd\u6027\u80fd\u5dee\u8ddd\uff1f<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nThis performance gap can be attributed to the domain shift<\/strong> and the more complex backgrounds present<\/strong> in split2 than in the simpler PlantVillage dataset (Fig. 2A) used for training<\/p>\n<\/blockquote>\n\n\n\n\u8bba\u6587\u660e\u786e\u6307\u51fa\uff0csplit2 \u5177\u6709\u6bd4 PlantVillage \u6570\u636e\u96c6\u66f4\u590d\u6742\u7684\u80cc\u666f\uff0c\u79cd\u7c7b\u4e5f\u66f4\u591a\u4e86\uff01<\/p>\n\n\n\nSplit1 \u53ea\u6709\u756a\u8304 \ud83c\udf45\uff0c\u800c split2 \u5305\u542b\u4e86\u82f9\u679c \ud83c\udf4e\u3001\u7389\u7c73 \ud83c\udf3d\u3001\u8461\u8404 \ud83c\udf47 \u7b49\u591a\u79cd\u4f5c\u7269\u3002<\/p>\n\n\n\n\u8bba\u6587\u5728 3.2.3 \u8282 \u4e2d\u5206\u6790\uff0c\u6b63\u662f\u56e0\u4e3a \u201c\u9886\u57df\u504f\u79fb\u201d\uff08domain shift\uff09 \u548c \u201c\u66f4\u590d\u6742\u7684\u80cc\u666f\u201d \u8fd9\u4e24\u5927\u56e0\u7d20\uff0c\u5bfc\u81f4\u4e86\u6a21\u578b\u5728 split2 \u4e0a\u7684\u6cdb\u5316\u6311\u6218\u3002<\/p>\n\n\n\n\u7b80\u5355\u6765\u8bf4\uff0c\u6a21\u578b\u5728\u80cc\u666f\u76f8\u5bf9\u7b80\u5355\u7684 PlantVillage \u6570\u636e\u96c6\u4e0a\u8bad\u7ec3\u540e\uff0c\u518d\u53bb\u8bc6\u522b\u80cc\u666f\u540c\u6837\u7b80\u5355\u7684\u756a\u8304\uff08split1\uff09\uff0c\u8868\u73b0\u5f88\u597d\u3002\u4f46\u662f\u5f53\u5b83\u9762\u5bf9\u80cc\u666f\u590d\u6742\u3001\u4f5c\u7269\u79cd\u7c7b\u53c8\u5b8c\u5168\u4e0d\u540c\u7684 split2 \u65f6\uff0c\u5c31\u9047\u5230\u4e86\u56f0\u96be\u3002<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nWhile split1 contains diseases with relatively more consistent features, split2 introduces additional variability that poses a challenge for models trained on simpler datasets. <\/p>\n<\/blockquote>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity is-style-wide\"\/>\n\n\n\n<h2 class=\"wp-block-heading has-text-align-center\">4.\u7ed3\u8bba\u548c\u672a\u6765\u5de5\u4f5c\u5c55\u671b<\/h2>\n\n\n\n<h2 class=\"wp-block-heading\">4.1\u7ed3\u8bba<\/h2>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nOur approach incorporates several key components: (1) a DCon-Adapter to enhance image feature representation, (2) a WDM to promote image-text interaction, and (3) the application of PlantCaFo and PlantCaFo* in practical scenarios demonstrates the effectiveness of the first two proposed methods.<\/p>\n\n\n\nTrans:\u6211\u4eec\u7684\u65b9\u6cd5\u5305\u542b\u4e09\u4e2a\u6838\u5fc3\u7ec4\u4ef6\uff1a(1)aDCon\u9002\u914d\u5668\u7528\u4e8e\u63d0\u5347\u56fe\u50cf\u7279\u5f81\u8868\u5f81\u8d28\u91cf\uff0c(2)WDM\u6280\u672f\u4fc3\u8fdb\u56fe\u50cf\u4e0e\u6587\u672c\u7684\u4ea4\u4e92\u4f5c\u7528\uff0c(3)\u5728\u5b9e\u9645\u573a\u666f\u4e2d\u5e94\u7528PlantCaFo\u548cPlantCaFo*\u9a8c\u8bc1\u4e86\u524d\u4e24\u79cd\u65b9\u6cd5\u7684\u6709\u6548\u6027\u3002<\/p>\n<\/blockquote>\n\n\n\n\u5728\u4f5c\u8005\u603b\u7ed3\u4e86\u4ed6\u4eec\u7684\u4e3b\u8981\u8d21\u732e\uff08\u4f8b\u5982 DCon-Adapter \u548c WDM \uff09\u4e4b\u540e\uff0c\u4ed6\u4eec\u63a5\u7740\u5728\u7b2c\u4e8c\u6bb5\u4e2d\u5766\u7387\u5730\u6307\u51fa\u4e86\u8fd9\u4e2a\u65b9\u6cd5\u7684\u201c\u5c40\u9650\u6027\u201d\uff08limitations\uff09\u3002<\/p>\n\n\n\n\u8fd9\u4e0e\u6211\u4eec\u521a\u624d\u8ba8\u8bba\u7684\u5728 split2 \u6570\u636e\u96c6\u4e0a\u7684\u6cdb\u5316\u6311\u6218\u662f\u5b8c\u5168\u4e00\u81f4\u7684\u3002\u6839\u636e\u8bba\u6587\u7684\u8bf4\u6cd5\uff0c\u5c3d\u7ba1 PlantCaFo \u5728\u53d7\u63a7\u73af\u5883\u4e2d\u8868\u73b0\u5f3a\u52b2\uff0c\u4f46\u5b83\u5728\u4ec0\u4e48\u60c5\u51b5\u4e0b\u80fd\u529b\u4f1a\u53d7\u5230\u9650\u5236\uff1f<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nHowever, there are certain limitations to our approach. While PlantCaFo demonstrates strong performance in controlled environments, its ability to generalize to highly diverse and complex agricultural scenarios may be limited<\/strong> because of the inherent challenges in handling variations in plant disease appearance<\/em> and image quality<\/em>. The use of the DCon-Adapter, while improving the feature extraction process, still faces difficulties in capturing all fine-grained disease patterns across different plant species. Additionally, although our approach works effectively on out-of-distribution datasets, the performance gap between different datasets, especially those with complex backgrounds or rare diseases, suggests that further improvements in model robustness are needed<\/strong>.<\/p>\n\n\n\nTrans:\u4e0d\u8fc7\uff0c\u6211\u4eec\u7684\u65b9\u6cd5\u4ecd\u5b58\u5728\u4e00\u4e9b\u5c40\u9650\u6027\u3002\u867d\u7136PlantCaFo\u5728\u53d7\u63a7\u73af\u5883\u4e2d\u8868\u73b0\u4f18\u5f02\uff0c\u4f46\u7531\u4e8e\u5904\u7406\u690d\u7269\u75c5\u5bb3\u5916\u89c2\u548c\u56fe\u50cf\u8d28\u91cf\u5dee\u5f02\u7684\u56fa\u6709\u6311\u6218\uff0c\u5176\u5728\u9ad8\u5ea6\u591a\u6837\u5316\u548c\u590d\u6742\u7684\u519c\u4e1a\u573a\u666f\u4e2d\u7684\u6cdb\u5316\u80fd\u529b\u53ef\u80fd\u53d7\u9650\u3002\u5c3d\u7ba1\u4f7f\u7528DCN\u9002\u914d\u5668\u6539\u8fdb\u4e86\u7279\u5f81\u63d0\u53d6\u8fc7\u7a0b\uff0c\u4f46\u5728\u6355\u6349\u4e0d\u540c\u690d\u7269\u7269\u79cd\u7684\u7ec6\u7c92\u5ea6\u75c5\u5bb3\u6a21\u5f0f\u65f6\u4ecd\u9762\u4e34\u56f0\u96be\u3002\u6b64\u5916\uff0c\u867d\u7136\u6211\u4eec\u7684\u65b9\u6cd5\u5728\u5206\u5e03\u5916\u6570\u636e\u96c6\u4e0a\u8868\u73b0\u826f\u597d\uff0c\u4f46\u4e0d\u540c\u6570\u636e\u96c6\uff08\u5c24\u5176\u662f\u5177\u6709\u590d\u6742\u80cc\u666f\u6216\u7f55\u89c1\u75c5\u5bb3\u7684\u6570\u636e\u96c6\uff09\u4e4b\u95f4\u7684\u6027\u80fd\u5dee\u8ddd\u8868\u660e\uff0c\u4ecd\u9700\u8fdb\u4e00\u6b65\u63d0\u5347\u6a21\u578b\u7684\u9c81\u68d2\u6027\u3002<\/p>\n<\/blockquote>\n\n\n\n\u8fd9\u6b63\u662f\u8bba\u6587\u5728\u7ed3\u8bba\u4e2d\u6307\u51fa\u7684\u6838\u5fc3\u5c40\u9650\u6027\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u590d\u6742\u573a\u666f\u7684\u6cdb\u5316\u80fd\u529b\u53d7\u9650\uff1a\u7279\u522b\u662f\u5728\u5904\u7406\u9ad8\u5ea6\u591a\u6837\u5316\u548c\u590d\u6742\u7684\u519c\u4e1a\u573a\u666f\u65f6 \u3002<\/li>\n\n\n\n<li>\u6570\u636e\u96c6\u4e4b\u95f4\u7684\u6027\u80fd\u5dee\u8ddd\uff1a\u5c31\u50cf\u6211\u4eec\u770b\u5230\u7684\uff0c\u6a21\u578b\u5728 PlantVillage \u548c PDL split2 \u4e0a\u7684\u8868\u73b0\u5dee\u5f02\u5f88\u5927 \u3002<\/li>\n<\/ol>\n\n\n\n\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e9b\u5c40\u9650\u6027\uff0c\u8bba\u6587\u5728\u7ed3\u8bba\u7684\u6700\u540e\u90e8\u5206\u63d0\u51fa\u4e86\u4e09\u4e2a\u201c\u672a\u6765\u5de5\u4f5c\u201d\u7684\u65b9\u5411 \u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">4.2\u672a\u6765\u5de5\u4f5c\u5c55\u671b<\/h2>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\nWe propose several potential directions for future work: (1)Designing hierarchical models:<\/strong> For complex plant disease recognition tasks, a hierarchical model architecture can be designed to classify plants and diseases at different levels. The first layer can perform coarse classification (e.g., plant type recognition), whereas the second layer can further identify specific diseases. (2) Designing specialized adapters for different plant disease categories:<\/strong> Future work could explore the design of multiple, task-specific adapters for plant disease recognition. By categorizing plant diseases into broader groups, distinct adapters can be tailored for each category, enabling the model to learn more specialized features. This modular approach may improve the performance on diverse disease types and enhance the model's ability to generalize across different categories. (3) Designing an adapter trained via meta-learning:<\/strong> By leveraging the concept of meta-learning, an adapter that can adapt quickly to few-shot tasks can be designed. Through training on multiple tasks, the meta-learning model can learn how to adjust the adapter's parameters more effectively, thereby demonstrating stronger adaptability and generalization abilities for new plant disease tasks.<\/p>\n\n\n\n\u6211\u4eec\u63d0\u51fa\u672a\u6765\u7814\u7a76\u7684\u82e5\u5e72\u6f5c\u5728\u65b9\u5411\uff1a(1)\u8bbe\u8ba1\u5c42\u6b21\u5316\u6a21\u578b\u67b6\u6784\uff1a\u9488\u5bf9\u590d\u6742\u7684\u690d\u7269\u75c5\u5bb3\u8bc6\u522b\u4efb\u52a1\uff0c\u53ef\u6784\u5efa\u5206\u5c42\u6a21\u578b\u67b6\u6784\uff0c\u5b9e\u73b0\u4e0d\u540c\u5c42\u7ea7\u7684\u690d\u7269\u4e0e\u75c5\u5bb3\u5206\u7c7b\u3002\u7b2c\u4e00\u5c42\u53ef\u8fdb\u884c\u7c97\u5206\u7c7b\uff08\u5982\u690d\u7269\u7c7b\u578b\u8bc6\u522b\uff09\uff0c\u7b2c\u4e8c\u5c42\u5219\u80fd\u8fdb\u4e00\u6b65\u8bc6\u522b\u7279\u5b9a\u75c5\u5bb3\u3002(2)\u5f00\u53d1\u9488\u5bf9\u4e0d\u540c\u75c5\u5bb3\u7c7b\u522b\u7684\u4e13\u7528\u9002\u914d\u5668\uff1a\u672a\u6765\u7814\u7a76\u53ef\u63a2\u7d22\u4e3a\u690d\u7269\u75c5\u5bb3\u8bc6\u522b\u8bbe\u8ba1\u591a\u4e2a\u4efb\u52a1\u4e13\u7528\u9002\u914d\u5668\u3002\u901a\u8fc7\u5c06\u690d\u7269\u75c5\u5bb3\u5f52\u7c7b\u4e3a\u66f4\u5e7f\u6cdb\u7684\u7c7b\u522b\uff0c\u53ef\u4e3a\u6bcf\u4e2a\u7c7b\u522b\u5b9a\u5236\u4e13\u5c5e\u9002\u914d\u5668\uff0c\u4f7f\u6a21\u578b\u80fd\u5b66\u4e60\u66f4\u4e13\u4e1a\u7684\u7279\u5f81\u3002\u8fd9\u79cd\u6a21\u5757\u5316\u8bbe\u8ba1\u6709\u671b\u63d0\u5347\u5bf9\u591a\u79cd\u75c5\u5bb3\u7c7b\u578b\u7684\u8bc6\u522b\u6027\u80fd\uff0c\u5e76\u589e\u5f3a\u6a21\u578b\u8de8\u7c7b\u522b\u6cdb\u5316\u80fd\u529b\u3002(3)\u6784\u5efa\u5143\u5b66\u4e60\u8bad\u7ec3\u7684\u9002\u914d\u5668\uff1a\u5229\u7528\u5143\u5b66\u4e60\u6982\u5ff5\uff0c\u53ef\u8bbe\u8ba1\u51fa\u80fd\u5feb\u901f\u9002\u5e94\u5c11\u6837\u672c\u4efb\u52a1\u7684\u9002\u914d\u5668\u3002\u901a\u8fc7\u591a\u4efb\u52a1\u8bad\u7ec3\uff0c\u5143\u5b66\u4e60\u6a21\u578b\u80fd\u66f4\u6709\u6548\u5730\u8c03\u6574\u9002\u914d\u5668\u53c2\u6570\uff0c\u4ece\u800c\u5728\u65b0\u578b\u690d\u7269\u75c5\u5bb3\u8bc6\u522b\u4efb\u52a1\u4e2d\u5c55\u73b0\u51fa\u66f4\u5f3a\u7684\u9002\u5e94\u6027\u548c\u6cdb\u5316\u80fd\u529b\u3002<\/p>\n<\/blockquote>\n\n\n\n<h2 class=\"wp-block-heading\">4.2.1\u8bbe\u8ba1\u5c42\u6b21\u5316\u6a21\u578b\u67b6\u6784<\/h2>\n\n\n\n\u8bba\u6587\u4e2d\u63d0\u5230\uff0c\u8fd9\u79cd\u67b6\u6784\u53ef\u4ee5\u5206\u5c42\u5206\u7c7b\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u7b2c\u4e00\u5c42 (Coarse)\uff1a\u5148\u6267\u884c\u7c97\u5206\u7c7b\uff08\u4f8b\u5982\uff0c\u8bc6\u522b\u690d\u7269\u7c7b\u578b\uff0c\u662f\u82f9\u679c\u8fd8\u662f\u756a\u8304\uff09\u3002<\/li>\n\n\n\n<li>\u7b2c\u4e8c\u5c42 (Fine)\uff1a\u518d\u8bc6\u522b\u7279\u5b9a\u75be\u75c5\uff08\u4f8b\u5982\uff0c\u662f\u65e9\u75ab\u75c5\u8fd8\u662f\u665a\u75ab\u75c5\uff09\u3002<\/li>\n<\/ul>\n\n\n\n\u56de\u60f3\u4e00\u4e0b\u6211\u4eec\u5728\u6cdb\u5316\u5b9e\u9a8c\uff08\u88686\uff09\u4e2d\u770b\u5230\u7684\u96be\u9898\uff1a\u5f53\u6a21\u578b\uff08\u5728\u756a\u8304\u3001\u571f\u8c46\u7b49\u4e0a\u8bad\u7ec3\uff09\u7a81\u7136\u9047\u5230\u4e00\u4e2a\u5168\u65b0\u7684\u7269\u79cd\uff08\u5982 split2 \u4e2d\u7684\u82f9\u679c\u6216\u7389\u7c73\uff09\u65f6\uff0c\u5b83\u7684\u8868\u73b0\u5c31\u4e0b\u964d\u4e86 \u3002<\/p>\n\n\n\n\u4f60\u8ba4\u4e3a\uff0c\u4e00\u4e2a\u201c\u5c42\u6b21\u5316\u201d\u7684\u6a21\u578b\uff0c\u5148\u628a\u4efb\u52a1\u5206\u89e3\u4e3a\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u8bc6\u522b\u8fd9\u662f\u201c\u82f9\u679c\u201d<\/li>\n\n\n\n<li>\u8bc6\u522b\u82f9\u679c\u5f97\u4e86\u201c\u4ec0\u4e48\u75c5\u201d<\/li>\n<\/ol>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8fd9\u79cd\u65b9\u5f0f\u662f\u5982\u4f55\u5e2e\u52a9\u89e3\u51b3\u90a3\u4e2a\u201c\u6cdb\u5316\u96be\u9898\u201d\u7684\u5462\uff1f\n<ul class=\"wp-block-list\">\n<li>\u89e3\u51b3\u4e86\u591a\u7269\u79cd\u60c5\u51b5\u4e0b\u8bc6\u522b\u4e0d\u51c6\u786e\u7684\u95ee\u9898<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n\u5b83\u4e4b\u6240\u4ee5\u80fd\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\uff0c\u662f\u56e0\u4e3a\u5b83\u628a\u4e00\u4e2a\u975e\u5e38\u590d\u6742\u7684\u201c\u4e00\u6b65\u5230\u4f4d\u201d\u7684\u4efb\u52a1\uff08\u6bd4\u5982\u572838\u4e2a\u7c7b\u522b\u4e2d\u540c\u65f6\u8bc6\u522b\u201c\u7269\u79cd\u201d\u548c\u201c\u75c5\u5bb3\u201d\uff09\u5206\u89e3\u6210\u4e86\u4e24\u4e2a\u66f4\u7b80\u5355\u3001\u66f4\u6e05\u6670\u7684\u4efb\u52a1 \uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u8fd9\u662f\u4ec0\u4e48\u690d\u7269\uff1f\uff08\u756a\u8304\uff1f\u82f9\u679c\uff1f\u8fd8\u662f\u7389\u7c73\uff1f\uff09<\/li>\n\n\n\n<li>\u8fd9\u4e2a\u690d\u7269\u5f97\u4e86\u4ec0\u4e48\u75c5\uff1f\uff08\u65e9\u75ab\u75c5\uff1f\u8fd8\u662f\u665a\u75ab\u75c5\uff1f\uff09<\/li>\n<\/ol>\n\n\n\n\u8fd9\u79cd\u201c\u5206\u800c\u6cbb\u4e4b\u201d\u7684\u7b56\u7565 \uff0c\u8ba9\u6a21\u578b\u53ef\u4ee5\u5148\u4e13\u6ce8\u4e8e\u8bc6\u522b\u201c\u690d\u7269\u7684\u6574\u4f53\u7279\u5f81\u201d\uff0c\u786e\u5b9a\u7269\u79cd\u540e\uff0c\u518d\u8c03\u7528\u4e00\u4e2a\u4e13\u95e8\u8bc6\u522b\u201c\u8be5\u7269\u79cd\u75c5\u5bb3\u7279\u5f81\u201d\u7684\u5b50\u6a21\u578b\u3002\u8fd9\u4f7f\u5f97\u6a21\u578b\u66f4\u52a0\u9c81\u68d2\uff0c\u4e5f\u66f4\u5bb9\u6613\u6cdb\u5316\u5230\u65b0\u7684\u7269\u79cd\u548c\u75c5\u5bb3\u4e0a\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity is-style-dots\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">4.2.2 \u5f00\u53d1\u9488\u5bf9\u4e0d\u540c\u75c5\u5bb3\u7c7b\u522b\u7684\u4e13\u7528\u9002\u914d\u5668<\/h2>\n\n\n\n\u8fd9\u4e2a\u60f3\u6cd5\u662f\uff0c\u672a\u6765\u7684\u7814\u7a76\u53ef\u4ee5\u63a2\u7d22\u8bbe\u8ba1\u591a\u4e2a\u3001\u9488\u5bf9\u7279\u5b9a\u4efb\u52a1\u7684\u9002\u914d\u5668\uff0c\u800c\u4e0d\u662f\u50cf\u73b0\u5728\u8fd9\u6837\u7528\u4e00\u4e2a\u9002\u914d\u5668\uff08DCon-Adapter\uff09\u6765\u5904\u7406\u6240\u6709\u7684\u75c5\u5bb3 \u3002<\/p>\n\n\n\n\u5177\u4f53\u7684\u5b9e\u73b0\u601d\u8def\u662f\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u5206\u7c7b\uff1a\u5148\u628a\u690d\u7269\u75c5\u5bb3\u5206\u6210\u51e0\u4e2a\u66f4\u5e7f\u6cdb\u7684\u7ec4\uff08broader groups\uff09 \u3002<\/li>\n\n\n\n<li>\u5b9a\u5236\uff1a\u7136\u540e\uff0c\u4e3a\u6bcf\u4e2a\u7ec4\u201c\u91cf\u8eab\u5b9a\u505a\u201d\u4e00\u4e2a\u4e13\u95e8\u7684\u9002\u914d\u5668 \u3002<\/li>\n<\/ol>\n\n\n\n\u4f5c\u8005\u8ba4\u4e3a\uff0c\u8fd9\u79cd\u201c\u6a21\u5757\u5316\u201d\u7684\u65b9\u6cd5\u6709\u4e24\u4e2a\u4e3b\u8981\u597d\u5904\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u80fd\u8ba9\u6a21\u578b\u4e3a\u6bcf\u4e2a\u7c7b\u522b\u5b66\u4e60\u5230\u66f4\u4e13\u4e1a\u7684\u7279\u5f81 \u3002<\/li>\n\n\n\n<li>\u53ef\u80fd\u4f1a\u63d0\u9ad8\u6a21\u578b\u5728\u5904\u7406\u591a\u79cd\u4e0d\u540c\u75be\u75c5\u7c7b\u578b\u65f6\u7684\u6027\u80fd\uff0c\u5e76\u589e\u5f3a\u5176\u5728\u4e0d\u540c\u7c7b\u522b\u95f4\u7684\u6cdb\u5316\u80fd\u529b \u3002<\/li>\n\n\n\n<li><\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity is-style-dots\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">4.2.3 \u6784\u5efa\u5143\u5b66\u4e60\u8bad\u7ec3\u9002\u914d\u5668<\/h2>\n\n\n\n\u201c\u5143\u5b66\u4e60\u201d\u7684\u6838\u5fc3\u601d\u60f3\u662f\u201c\u5b66\u4f1a\u5982\u4f55\u5b66\u4e60\u201d\u3002<\/p>\n\n\n\n\u5728\u8fd9\u7bc7\u8bba\u6587\u7684\u8bbe\u60f3\u4e2d\uff0c\u4e0d\u662f\u50cf\u73b0\u5728\u8fd9\u6837\u5728\u4e00\u4e2a\u201c\u5927\u201d\u4efb\u52a1\u4e0a\uff08\u6bd4\u5982 PlantVillage\uff09\u8bad\u7ec3 DCon-Adapter\uff0c\u800c\u662f\u8ba9\u6a21\u578b\u5728\u201c\u5927\u91cf\u3001\u4e0d\u540c\u201d\u7684\u5c11\u6837\u672c\u4efb\u52a1\u4e0a\u8fdb\u884c\u8bad\u7ec3 \u3002<\/p>\n\n\n\n\u4f8b\u5982\uff0c\u6a21\u578b\u4f1a\u7ec3\u4e60\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4efb\u52a11\uff1a\u4ece2\u5f20\u201c\u82f9\u679c\u9ed1\u661f\u75c5\u201d\u548c2\u5f20\u201c\u82f9\u679c\u9508\u75c5\u201d\u56fe\u7247\u4e2d\u5b66\u4f1a\u533a\u5206\u5b83\u4eec\u3002<\/li>\n\n\n\n<li>\u4efb\u52a12\uff1a\u4ece2\u5f20\u201c\u7389\u7c73\u9508\u75c5\u201d\u548c2\u5f20\u201c\u7389\u7c73\u7070\u6591\u75c5\u201d\u56fe\u7247\u4e2d\u5b66\u4f1a\u533a\u5206\u5b83\u4eec\u3002<\/li>\n\n\n\n<li>\u4efb\u52a13\uff1a\u2026\u2026\uff08\u4ee5\u6b64\u7c7b\u63a8\uff0c\u6210\u767e\u4e0a\u5343\u4e2a\u5c0f\u4efb\u52a1\uff09<\/li>\n<\/ul>\n\n\n\n\u901a\u8fc7\u5728\u8fd9\u4e48\u591a\u4e0d\u540c\u7684\u201c\u5c11\u6837\u672c\u4efb\u52a1\u201d\u4e0a\u8fdb\u884c\u8bad\u7ec3\uff0c\u8fd9\u4e2a\u9002\u914d\u5668\u4f1a\u53d8\u5f97\u975e\u5e38\u64c5\u957f\u201c\u5feb\u901f\u9002\u5e94\u201d \u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4e0e\u6211\u4eec\u5f53\u524d\u7814\u7a76\u7684 DCon-Adapter\uff08\u5728\u4e00\u4e2a\u4efb\u52a1\u4e0a\u8bad\u7ec3\uff09\u76f8\u6bd4\uff0c\u4f60\u8ba4\u4e3a\u8fd9\u79cd\u201c\u5143\u5b66\u4e60\u201d\u8bad\u7ec3\u51fa\u6765\u7684\u9002\u914d\u5668\uff0c\u5728\u9047\u5230\u4e00\u79cd\u5168\u65b0\u7684\u690d\u7269\u75c5\u5bb3\uff08\u6bd4\u5982\u4e00\u79cd\u5b83\u4ece\u672a\u89c1\u8fc7\u7684\u8c46\u7c7b\u75c5\u5bb3\uff09\u65f6\uff0c\u4f1a\u6709\u4ec0\u4e48\u4f18\u52bf\uff1f\n<ul class=\"wp-block-list\">\n<li>\u80fd\u8fc5\u901f\u5c06\u8fd9\u4e2a\u75c5\u5bb3\u5224\u65ad\u6210\u65b0\u7684\u4e00\u7c7b<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n\u6211\u4eec\u5f53\u524d\u7684 DCon-Adapter \u64c5\u957f\u8bc6\u522b\u5b83\u5df2\u7ecf\u5b66\u8fc7<\/em>\u7684\u75c5\u5bb3\u3002\u800c\u201c\u5143\u5b66\u4e60\u201d\u9002\u914d\u5668\uff08\u5728\u8bb8\u591a\u4efb\u52a1\u4e0a\u8bad\u7ec3\u8fc7\uff09\u7684\u6838\u5fc3\u4f18\u52bf\u5728\u4e8e\u5b83\u201c\u5b66\u4f1a\u4e86\u5982\u4f55\u5b66\u4e60\u201d\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6839\u636e\u8bba\u6587 \uff0c\u5f53\u5b83\u9047\u5230\u4e00\u4e2a\u5168\u65b0\u7684\u75c5\u5bb3\u65f6\uff0c\u8fd9\u79cd\u201c\u5b66\u4f1a\u5b66\u4e60\u201d\u7684\u80fd\u529b\u8ba9\u5b83\u5728\u201c\u9002\u5e94\u201d\u8fd9\u4e2a\u65b0\u4efb\u52a1\u65f6\u6709\u4ec0\u4e48\u5173\u952e\u4f18\u52bf\uff1f\n<ul class=\"wp-block-list\">\n<li>\u5c55\u73b0\u51fa\u66f4\u5f3a\u7684\u9002\u5e94\u6027\u548c\u6cdb\u5316\u80fd\u529b<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n\u8fd9\u6b63\u662f\u201c\u5143\u5b66\u4e60\u201d\u7684\u6838\u5fc3\u4f18\u52bf\uff1a\u901a\u8fc7\u5728\u5927\u91cf\u4e0d\u540c\u4efb\u52a1\u4e0a\u201c\u5b66\u4f1a\u5982\u4f55\u5b66\u4e60\u201d\uff0c\u5b83\u5728\u9762\u5bf9\u5168\u65b0\u7684\u75c5\u5bb3\u4efb\u52a1\u65f6\uff0c\u80fd\u591f\u5c55\u73b0\u51fa\u66f4\u5f3a\u7684\u9002\u5e94\u80fd\u529b\u548c\u6cdb\u5316\u80fd\u529b \u3002<\/p>\n","protected":false},"excerpt":{"rendered":"\u672c\u671f\u8bc4\u8ff0\u6587\u732e\uff1a 1.\u7814\u7a76\u80cc\u666f\u4e0e\u95ee\u9898 \u8fd9\u7bc7\u8bba\u6587\u7684\u5f15\u8a00\uff08Introduction\uff09\u91cc\u63d0\u5230\uff0c\u81ea\u52a8\u8bc6\u522b\u690d\u7269\u75c5\u5bb3\u5bf9\u4e8e\u7cae\u98df […]<\/p>\n","protected":false},"author":1,"featured_media":3773,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[15],"tags":[83],"class_list":["post-3771","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-microtalk","tag-zuhui4"],"_links":{"self":[{"href":"https:\/\/www.leexinghai.com\/aic\/wp-json\/wp\/v2\/posts\/3771","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.leexinghai.com\/aic\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.leexinghai.com\/aic\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.leexinghai.com\/aic\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.leexinghai.com\/aic\/wp-json\/wp\/v2\/comments?post=3771"}],"version-history":[{"count":16,"href":"https:\/\/www.leexinghai.com\/aic\/wp-json\/wp\/v2\/posts\/3771\/revisions"}],"predecessor-version":[{"id":3808,"href":"https:\/\/www.leexinghai.com\/aic\/wp-json\/wp\/v2\/posts\/3771\/revisions\/3808"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.leexinghai.com\/aic\/wp-json\/wp\/v2\/media\/3773"}],"wp:attachment":[{"href":"https:\/\/www.leexinghai.com\/aic\/wp-json\/wp\/v2\/media?parent=3771"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.leexinghai.com\/aic\/wp-json\/wp\/v2\/categories?post=3771"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.leexinghai.com\/aic\/wp-json\/wp\/v2\/tags?post=3771"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}