Scholarlink

<p>The dual-branch hybrid model successfully combines the advantages of two models in object classification. Therefore, the study proposes a model named Explainable dual-branch hybrid mode (E-DUALNet) to take advantage of global and local features from two models, EfficientNetV2S and MobileNets. The research uses Explainable artificial intelligence (XAI) to provide insights into the image regions that contributed most to the model’s predictions. The proposed model’s ability to accurately classify each disease class is evaluated using Focus Score, a saliency metric. Besides, the research proposed improving the Focus Score to assess results with multiple images. By combining the complementary advantages of the two models, E-DUALNet achieves an accuracy of 97.98%, an F1-Score of 96.69%, and a quadratic weighted kappa of 97.31%, all of which are higher than the baseline models used for comparison. The proposed model also has good recognition performance and high reliability. These results showcase the advantages of using XAI in dual-branch hybrid models and the applicability of the proposed field of image processing in agriculture.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Explainable Dual-Branch Hybrid Model for Maize Leaf Disease Classification

  • 期刊论文

  • 发表日期：

    2025年6月25日

查看全文
  • Luyl-Da Quach,
  • Nguyen Thai-Nghe,
  • Chi-Ngon Nguyen

摘要

The dual-branch hybrid model successfully combines the advantages of two models in object classification. Therefore, the study proposes a model named Explainable dual-branch hybrid mode (E-DUALNet) to take advantage of global and local features from two models, EfficientNetV2S and MobileNets. The research uses Explainable artificial intelligence (XAI) to provide insights into the image regions that contributed most to the model’s predictions. The proposed model’s ability to accurately classify each disease class is evaluated using Focus Score, a saliency metric. Besides, the research proposed improving the Focus Score to assess results with multiple images. By combining the complementary advantages of the two models, E-DUALNet achieves an accuracy of 97.98%, an F1-Score of 96.69%, and a quadratic weighted kappa of 97.31%, all of which are higher than the baseline models used for comparison. The proposed model also has good recognition performance and high reliability. These results showcase the advantages of using XAI in dual-branch hybrid models and the applicability of the proposed field of image processing in agriculture.