Scholarlink

<p>A&#xa0;novel technique has been proposed for diagnosing mango leaf disease. Advanced image processing, feature extraction, and deep learning models have been implemented to facilitate rapid disease diagnosis, thereby fulfilling the above purpose. The proposed model is developed based on the Perona–Malik filter, Tensor Empirical Wavelet Transform (T-EWT), and Gaussian Copula Entropy. The Perona–Malik filter removes noise, while T‑EWT decomposes images into intrinsic sub-bands. Then, deep features are extracted using DenseNet-121 and DenseNet-201, and the resulting features are concatenated into a&#xa0;final feature vector. Gaussian Copula Entropy is then used for feature selection, and selected features are fed into the random forest. Model interpretability is assessed using Grad-CAM and an ablation study. The classification model achieved an indicative accuracy of 98.46%, with 98% sensitivity and 98.75% specificity. A&#xa0;five-fold cross-validation is employed for model validation using splitting techniques. This model yields effective results for the rapid, accurate diagnosis of mango leaves, ensuring sustainable harvesting performance.</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

Mango (Mangifera indica) Leaf Disease Prognosis Using Perona–Malik Filter and Gaussian Copula Entropy in Tensor Empirical Wavelet Transform

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • Rajneesh Kumar Patel,
  • Siddharth Singh Chouhan,
  • Harshlata Vishwakarma

摘要

A novel technique has been proposed for diagnosing mango leaf disease. Advanced image processing, feature extraction, and deep learning models have been implemented to facilitate rapid disease diagnosis, thereby fulfilling the above purpose. The proposed model is developed based on the Perona–Malik filter, Tensor Empirical Wavelet Transform (T-EWT), and Gaussian Copula Entropy. The Perona–Malik filter removes noise, while T‑EWT decomposes images into intrinsic sub-bands. Then, deep features are extracted using DenseNet-121 and DenseNet-201, and the resulting features are concatenated into a final feature vector. Gaussian Copula Entropy is then used for feature selection, and selected features are fed into the random forest. Model interpretability is assessed using Grad-CAM and an ablation study. The classification model achieved an indicative accuracy of 98.46%, with 98% sensitivity and 98.75% specificity. A five-fold cross-validation is employed for model validation using splitting techniques. This model yields effective results for the rapid, accurate diagnosis of mango leaves, ensuring sustainable harvesting performance.