Scholarlink

<p>Over the past two decades, climate change has severely impacted agriculture, affecting all stages of production. Farmers are increasingly adopting advanced technologies to mitigate these effects. This study introduces a&#xa0;novel deep learning model, PL-DenseNet, for accurately classifying pear leaf diseases. PL-DenseNet enhances the original DenseNet architecture by replacing the classification layer with three additional layers: Global Average Pooling 2D, Batch Normalization, and Dropout. The final layer includes four nodes dedicated to pear leaf disease classification. To improve accuracy and robustness, the model leverages transfer learning and data augmentation to expand the training data’s diversity and quantity. The model was trained on the DiaMOS Plant dataset, which contains 7337 images spanning four classes of pear leaf diseases, captured under real-field conditions with complex backgrounds, varying brightness, and disease similarity. Experimental results show that the proposed PL-DenseNet model outperforms baseline and state-of-the-art models, including EfficientNetB0, InceptionV2, MobileNetV2, ResNet50, and VGG16. It achieves an impressive accuracy of 99.18%, precision of 98.83%, recall of 99.06%, F1-score of 93.58, and a&#xa0;loss rate between 0.02 and 0.05. These findings highlight PL-DenseNet’s superior performance in disease classification and demonstrate the effectiveness of transfer learning and data augmentation techniques.</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

Optimizing Pear Leaf Disease Detection Through PL-DenseNet

  • 期刊论文

  • 发表日期：

    2025年2月11日

查看全文
  • Yonis Gulzar,
  • Zeynep Ünal

摘要

Over the past two decades, climate change has severely impacted agriculture, affecting all stages of production. Farmers are increasingly adopting advanced technologies to mitigate these effects. This study introduces a novel deep learning model, PL-DenseNet, for accurately classifying pear leaf diseases. PL-DenseNet enhances the original DenseNet architecture by replacing the classification layer with three additional layers: Global Average Pooling 2D, Batch Normalization, and Dropout. The final layer includes four nodes dedicated to pear leaf disease classification. To improve accuracy and robustness, the model leverages transfer learning and data augmentation to expand the training data’s diversity and quantity. The model was trained on the DiaMOS Plant dataset, which contains 7337 images spanning four classes of pear leaf diseases, captured under real-field conditions with complex backgrounds, varying brightness, and disease similarity. Experimental results show that the proposed PL-DenseNet model outperforms baseline and state-of-the-art models, including EfficientNetB0, InceptionV2, MobileNetV2, ResNet50, and VGG16. It achieves an impressive accuracy of 99.18%, precision of 98.83%, recall of 99.06%, F1-score of 93.58, and a loss rate between 0.02 and 0.05. These findings highlight PL-DenseNet’s superior performance in disease classification and demonstrate the effectiveness of transfer learning and data augmentation techniques.