Scholarlink

<p>Soybean growth is highly susceptible to Soybean Mosaic Virus (SMV), making early diagnosis essential for effective prevention and control. To enhance the accuracy of early diagnosis of Soybean Mosaic Virus, this study employed a&#xa0;hyperspectral camera to monitor and photograph infected soybean leaves for 11&#xa0;days prior to symptom manifestation, thereby constructing an early disease dataset for SMV. A&#xa0;hybrid detection model integrating Convolutional Neural Network (CNN) with an attention mechanism and Long Short-Term Memory (LSTM) was proposed, achieving an accuracy of 89.62% in detecting healthy and infected leaves, which is significantly higher than the 64.37% accuracy obtained with RGB images. Compared to single algorithms 1‑Dimensional Convolutional Neural Network (1D-CNN), LSTM, Support Vector Machine (SVM), and the combined algorithm CNN-LSTM, the improvements in accuracy were 5.56, 1.89, 7.3, and 0.7%, respectively. Notably, the model achieved a&#xa0;detection accuracy of 90.74% on the leaves at 2&#xa0;days post-inoculation, with subsequent results remaining relatively stable, indicating robust early diagnostic capabilities of this method. For m further practical application, a&#xa0;convenient diagnostic platform, “SMV-GPT” was developed using Open AI’s Chat GPT API to provide disease detection and management recommendations.</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

Hyperspectral Detection of Early Soybean Mosaic Disease Based On Deep Learning

  • 期刊论文

  • 发表日期：

    2025年6月17日

查看全文
  • Ziyan Zong,
  • Xuetong Zhai,
  • Hongfei Zhu,
  • Kai Huang,
  • Zhongzhi Han,
  • HeXiang Luan,
  • Tao Luan

摘要

Soybean growth is highly susceptible to Soybean Mosaic Virus (SMV), making early diagnosis essential for effective prevention and control. To enhance the accuracy of early diagnosis of Soybean Mosaic Virus, this study employed a hyperspectral camera to monitor and photograph infected soybean leaves for 11 days prior to symptom manifestation, thereby constructing an early disease dataset for SMV. A hybrid detection model integrating Convolutional Neural Network (CNN) with an attention mechanism and Long Short-Term Memory (LSTM) was proposed, achieving an accuracy of 89.62% in detecting healthy and infected leaves, which is significantly higher than the 64.37% accuracy obtained with RGB images. Compared to single algorithms 1‑Dimensional Convolutional Neural Network (1D-CNN), LSTM, Support Vector Machine (SVM), and the combined algorithm CNN-LSTM, the improvements in accuracy were 5.56, 1.89, 7.3, and 0.7%, respectively. Notably, the model achieved a detection accuracy of 90.74% on the leaves at 2 days post-inoculation, with subsequent results remaining relatively stable, indicating robust early diagnostic capabilities of this method. For m further practical application, a convenient diagnostic platform, “SMV-GPT” was developed using Open AI’s Chat GPT API to provide disease detection and management recommendations.