Scholarlink

<p>Accurate identification of plant diseases is paramount for ensuring optimal crop health and minimizing yield losses. This study focuses on <i>Solanum melongena</i> L. (eggplant), a widely cultivated vegetable highly susceptible to fungal infections. We employ a semantic segmentation approach to detect leaf symptoms and damage, utilizing deep learning models. A comparative analysis is conducted on three convolutional neural network architectures: U2-Net, U-Net, and WU-Net, for image segmentation tasks. Each model is trained on a dataset comprising 500 augmented images derived from an initial set of 200 images, with a resolution of 256 × 256 × 3. Model performance is evaluated based on pixel accuracy, Dice coefficient, and Intersection over Union. Experimental results demonstrate that U2-Net exhibits outstanding performance, particularly in capturing fine-grained details, attributed to its deeper architecture and enhanced feature extraction capabilities. The proposed U2-Net model achieved a training accuracy of 96% and test accuracy of 93%, demonstrating its effectiveness in precise symptom detection and plant disease classification. This study contributes to leaf disease detection, facilitating timely intervention and targeted disease management 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

Enhanced fungal disease detection in brinjal leaves using deep semantic segmentation with U2-Net

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月3日

查看全文
  • Ruchita R Patil,
  • Shwetha V,
  • Annamalai Muthusamy

摘要

Accurate identification of plant diseases is paramount for ensuring optimal crop health and minimizing yield losses. This study focuses on Solanum melongena L. (eggplant), a widely cultivated vegetable highly susceptible to fungal infections. We employ a semantic segmentation approach to detect leaf symptoms and damage, utilizing deep learning models. A comparative analysis is conducted on three convolutional neural network architectures: U2-Net, U-Net, and WU-Net, for image segmentation tasks. Each model is trained on a dataset comprising 500 augmented images derived from an initial set of 200 images, with a resolution of 256 × 256 × 3. Model performance is evaluated based on pixel accuracy, Dice coefficient, and Intersection over Union. Experimental results demonstrate that U2-Net exhibits outstanding performance, particularly in capturing fine-grained details, attributed to its deeper architecture and enhanced feature extraction capabilities. The proposed U2-Net model achieved a training accuracy of 96% and test accuracy of 93%, demonstrating its effectiveness in precise symptom detection and plant disease classification. This study contributes to leaf disease detection, facilitating timely intervention and targeted disease management in agriculture.