Scholarlink

This study introduces a novel approach to potato leaf disease detection using a DenseNet121-based convolutional neural network (CNN) model that achieves remarkable accuracy and efficiency. By utilizing a carefully curated dataset comprising six categories: early blight, late blight, healthy, virus, fungi, and pest and employing advanced preprocessing and augmentation techniques, the model demonstrated exceptional robustness and generalization. The proposed system achieved an impressive testing accuracy of 97.40%, precision of 97.10%, recall of 97.19%, and F1-score of 97.12%. Additionally, the model showed efficient performance, with an inference time of 0.1847 s per batch and an average inference time of 0.0029 s per image, highlighting its suitability for real-time applications. The optimized training pipeline effectively extracts complex patterns and differentiates subtle variations between visually similar categories, thereby ensuring high-confidence predictions. This scalable, real-time solution bridges the gap between traditional manual diagnostics and modern agricultural needs, and offers a practical tool for timely and accurate disease management. By addressing critical challenges in potato farming, such as yield losses and quality deterioration, this study establishes a new standard for automated plant disease diagnostics. It promises to revolutionize agricultural practices by fostering efficiency, sustainability, and equitable access to advanced technological solutions.

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

Harnessing Convolutional Neural Networks for Potato Leaf Disease Detection: A Proposed Model

  • 章节

  • 发表日期：

    2025年

查看全文
  • Mir Maruf Ahmed,
  • Rakin Sad Aftab,
  • Sultanul Arifeen Hamim,
  • Md. Abdullah-Al-Jubair,
  • Dip Nandi

摘要

This study introduces a novel approach to potato leaf disease detection using a DenseNet121-based convolutional neural network (CNN) model that achieves remarkable accuracy and efficiency. By utilizing a carefully curated dataset comprising six categories: early blight, late blight, healthy, virus, fungi, and pest and employing advanced preprocessing and augmentation techniques, the model demonstrated exceptional robustness and generalization. The proposed system achieved an impressive testing accuracy of 97.40%, precision of 97.10%, recall of 97.19%, and F1-score of 97.12%. Additionally, the model showed efficient performance, with an inference time of 0.1847 s per batch and an average inference time of 0.0029 s per image, highlighting its suitability for real-time applications. The optimized training pipeline effectively extracts complex patterns and differentiates subtle variations between visually similar categories, thereby ensuring high-confidence predictions. This scalable, real-time solution bridges the gap between traditional manual diagnostics and modern agricultural needs, and offers a practical tool for timely and accurate disease management. By addressing critical challenges in potato farming, such as yield losses and quality deterioration, this study establishes a new standard for automated plant disease diagnostics. It promises to revolutionize agricultural practices by fostering efficiency, sustainability, and equitable access to advanced technological solutions.