Scholarlink

<p>The ability of deep neural networks to extract highly transferable features has significantly improved image recognition tasks, including plant disease detection. However, most real-world datasets contain unlabeled or partially labeled data, making it difficult for deep models to learn effectively. To address this challenge, this paper proposes a deep dynamic joint adaptation network (DDJAN) for plant disease recognition, which leverages transfer learning and entropy minimization to increase classification performance. Unlike traditional domain adaptation methods such as deep domain confusion (DDC), deep adaptation networks (DANs), and joint adaptation networks (JANs), DDJAN dynamically adjusts the importance of marginal and conditional distributions, ensuring better adaptation to domain shifts. The proposed method was evaluated on plant disease datasets and achieved recognition accuracies of 97.30%, 95.05%, and 94.60%, outperforming the JAN (88.80%) and other baseline approaches. The results demonstrate that DDJAN effectively reduces classification uncertainty and improves generalization in cross-domain plant disease recognition. These findings highlight the potential of DDJAN as a robust and adaptive solution for agricultural disease diagnosis, with future work focusing on integrating transformer-based architectures for further enhancements.</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

Plant leaf disease recognition using deep dynamic joint adaptation networks based on partially labeled data

  • 期刊论文

  • 发表日期：

    2025年5月26日

查看全文
  • Isack Bulugu

摘要

The ability of deep neural networks to extract highly transferable features has significantly improved image recognition tasks, including plant disease detection. However, most real-world datasets contain unlabeled or partially labeled data, making it difficult for deep models to learn effectively. To address this challenge, this paper proposes a deep dynamic joint adaptation network (DDJAN) for plant disease recognition, which leverages transfer learning and entropy minimization to increase classification performance. Unlike traditional domain adaptation methods such as deep domain confusion (DDC), deep adaptation networks (DANs), and joint adaptation networks (JANs), DDJAN dynamically adjusts the importance of marginal and conditional distributions, ensuring better adaptation to domain shifts. The proposed method was evaluated on plant disease datasets and achieved recognition accuracies of 97.30%, 95.05%, and 94.60%, outperforming the JAN (88.80%) and other baseline approaches. The results demonstrate that DDJAN effectively reduces classification uncertainty and improves generalization in cross-domain plant disease recognition. These findings highlight the potential of DDJAN as a robust and adaptive solution for agricultural disease diagnosis, with future work focusing on integrating transformer-based architectures for further enhancements.