Scholarlink

<p>A wide range of plant diseases that impact economically and nutritionally significant crops such as Beans, Cauliflower, Jackfruit, Malabar Spinach, and Mango pose a serious danger to Bangladesh’s agricultural output. Scalable, quick, and precise solutions are needed because traditional disease inspection techniques are lengthy and frequently out of reach for rural farmers. Based on a modified MobileNetV3Large architecture, this study suggests a lightweight, portable, and effective framework for detecting plant diseases. The model was trained on a hybrid dataset consisting of 26 classes that combined field-collected images from Bangladesh with publicly available sources, utilizing transfer learning and fine-tuning techniques to ensure robustness against real-world conditions such as changing lighting and background clutter. In order to improve model generalization and reduce overfitting, data augmentation techniques such as random flipping, rotation, brightness and contrast change, and zooming were utilized. Under identical experimental conditions, the performance of the suggested model was methodically compared to a customized Convolutional Neural Network, baseline MobileNetV3Large, and EfficientNetB1 utilizing metrics such as test accuracy, confusion matrix, precision, recall, F1-score, ROC-AUC, and precision-recall curves. Compared with the baseline MobileNetV3Large (98.97%), EfficientNetB1 (98.3%), and the custom Convolutional Neural Network (95.51%), the customized MobileNetV3Large demonstrated better predictive capability and generalizability across all classes, achieving superior test accuracy (99.37%). The outcomes confirm the system’s excellent diagnostic accuracy and viability for immediate use in mobile-based digital agriculture systems in rural regions with limited resources.</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

A customized MobileNetV3Large-based deep learning framework for plant disease detection

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月11日

查看全文
  • Jabin Rahaman,
  • Priyanti Paul,
  • Aditta Chowdhury,
  • Muhammad Quamruzzaman

摘要

A wide range of plant diseases that impact economically and nutritionally significant crops such as Beans, Cauliflower, Jackfruit, Malabar Spinach, and Mango pose a serious danger to Bangladesh’s agricultural output. Scalable, quick, and precise solutions are needed because traditional disease inspection techniques are lengthy and frequently out of reach for rural farmers. Based on a modified MobileNetV3Large architecture, this study suggests a lightweight, portable, and effective framework for detecting plant diseases. The model was trained on a hybrid dataset consisting of 26 classes that combined field-collected images from Bangladesh with publicly available sources, utilizing transfer learning and fine-tuning techniques to ensure robustness against real-world conditions such as changing lighting and background clutter. In order to improve model generalization and reduce overfitting, data augmentation techniques such as random flipping, rotation, brightness and contrast change, and zooming were utilized. Under identical experimental conditions, the performance of the suggested model was methodically compared to a customized Convolutional Neural Network, baseline MobileNetV3Large, and EfficientNetB1 utilizing metrics such as test accuracy, confusion matrix, precision, recall, F1-score, ROC-AUC, and precision-recall curves. Compared with the baseline MobileNetV3Large (98.97%), EfficientNetB1 (98.3%), and the custom Convolutional Neural Network (95.51%), the customized MobileNetV3Large demonstrated better predictive capability and generalizability across all classes, achieving superior test accuracy (99.37%). The outcomes confirm the system’s excellent diagnostic accuracy and viability for immediate use in mobile-based digital agriculture systems in rural regions with limited resources.