Scholarlink

<p>Apples are globally cultivated fruits known for their high nutritional value; however, plant diseases significantly affect yield and fruit quality, leading to economic losses. The integration of artificial intelligence in agriculture has enhanced disease detection through advanced deep learning techniques. In this study, a&#xa0;convolutional neural network-based model, PomaNet, is proposed for apple leaf disease detection and is compared with eight pre-trained models, namely, VGG16, ResNet50, DenseNet121, Xception, EfficientNetB3, InceptionV3, MobileNet, and AlexNet. Image preprocessing techniques, including resizing, normalization, Gaussian blurring, and contrast limited adaptive histogram equalization, are applied to improve dataset quality, while data augmentation is used to address class imbalance. Experimental results show that PomaNet outperforms the other models, achieving an accuracy of 99.42%, precision of 99.49%, recall of 99.46%, and F1&#xa0;score of 99.47%, along with Cohen’s kappa and Matthews correlation coefficient value of 99.27%. Additionally, low error values (mean absolute error: 0.0102, mean squared error: 0.0237, root mean squared error: 0.1540) confirm the effectiveness of the approach. The proposed framework demonstrates strong potential for real-time apple leaf disease detection, enabling timely intervention and improved crop management.</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

PomaNet: An Optimized CNN Framework for Apple Leaf Disease Detection Using Advanced Image Processing

  • 期刊论文

  • 发表日期：

    2026年5月19日

查看全文
  • Komalpreet Kaur,
  • Hardeep Kaur

摘要

Apples are globally cultivated fruits known for their high nutritional value; however, plant diseases significantly affect yield and fruit quality, leading to economic losses. The integration of artificial intelligence in agriculture has enhanced disease detection through advanced deep learning techniques. In this study, a convolutional neural network-based model, PomaNet, is proposed for apple leaf disease detection and is compared with eight pre-trained models, namely, VGG16, ResNet50, DenseNet121, Xception, EfficientNetB3, InceptionV3, MobileNet, and AlexNet. Image preprocessing techniques, including resizing, normalization, Gaussian blurring, and contrast limited adaptive histogram equalization, are applied to improve dataset quality, while data augmentation is used to address class imbalance. Experimental results show that PomaNet outperforms the other models, achieving an accuracy of 99.42%, precision of 99.49%, recall of 99.46%, and F1 score of 99.47%, along with Cohen’s kappa and Matthews correlation coefficient value of 99.27%. Additionally, low error values (mean absolute error: 0.0102, mean squared error: 0.0237, root mean squared error: 0.1540) confirm the effectiveness of the approach. The proposed framework demonstrates strong potential for real-time apple leaf disease detection, enabling timely intervention and improved crop management.