Scholarlink

<p>Training deep learning models for plant disease classification on small size datasets with heterogeneous backgrounds remains challenging issue. Proposed study overcomes the major limitations and weaknesses of thexisting contemporary methods. This study introduces a novel three-phase approach: (1) segmentation of disease regions using a Min–Max Hue Histogram-based technique to isolate the disease region with the most important data, (2) a lightweight, sequential deep learning model (PDCNet) trained from scratch to classify the diseases and (3) a disease recovery module to provide recommendation to farmer. Unlikexisting works, we avoid data augmentation and transfer learning toliminate overfitting risks and domain mismatches. Additionally, we created the Plant Disease Small Dataset (PDSD)—a new, realistic subset of the PlantVillage dataset featuring images with variednvironmental conditions and backgrounds.experimental results show that our method achieves an average classification accuracy of 91%, which is better than popular CNN architectures (VGG19 and ResNet50). This complete system provides an accurate,ffective, and convenient way for farmers to diagnose plant diseases and to receive recovery strategy, which can help promote the precision agriculture practices.</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

An effective approach for classification of plant disease of small size dataset by using sequential deep learning model with Min–Max Hue histogram based segmentation

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月29日

查看全文
  • Vijay Kumar Trivedi,
  • Piyush Kumar Shukla,
  • Anjana Pandey,
  • Noha Alduaiji,
  • J. Shreyas

摘要

Training deep learning models for plant disease classification on small size datasets with heterogeneous backgrounds remains challenging issue. Proposed study overcomes the major limitations and weaknesses of thexisting contemporary methods. This study introduces a novel three-phase approach: (1) segmentation of disease regions using a Min–Max Hue Histogram-based technique to isolate the disease region with the most important data, (2) a lightweight, sequential deep learning model (PDCNet) trained from scratch to classify the diseases and (3) a disease recovery module to provide recommendation to farmer. Unlikexisting works, we avoid data augmentation and transfer learning toliminate overfitting risks and domain mismatches. Additionally, we created the Plant Disease Small Dataset (PDSD)—a new, realistic subset of the PlantVillage dataset featuring images with variednvironmental conditions and backgrounds.experimental results show that our method achieves an average classification accuracy of 91%, which is better than popular CNN architectures (VGG19 and ResNet50). This complete system provides an accurate,ffective, and convenient way for farmers to diagnose plant diseases and to receive recovery strategy, which can help promote the precision agriculture practices.