Scholarlink

<p>Plant diseases significantly impact agricultural productivity, necessitating timely and accurate diagnosis. While existing deep learning-based systems offer promising solutions, they often lack robustness, struggle with unseen crop types and diseases, and are generally limited to single-disease detection using leaf imagery alone. This study proposes a comprehensive multi-class plant disease recognition framework to address these limitations. A generalized deep learning model based on the Small Inception architecture was developed to classify healthy and diseased leaves across diverse crops by focusing on localized lesion regions. The approach utilizes a patch-based analysis strategy to detect multiple diseases on a single leaf while removing crop-specific dependencies. Experimental results on the PlantVillage dataset demonstrate superior performance: the Small Inception model achieved a training accuracy of 91.81% with a loss of 0.2385, and a testing accuracy of 90.57% with a loss of 0.2954. In comparison, MiniVGGNet and LeNet5 achieved lower testing accuracies of 80.24% and 80.72%, respectively. These results affirm the robustness and generalizability of the proposed system, which supports accurate multi-disease detection and extends applicability to diverse crop types and real-world farming conditions.</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

Deep Learning-Based Multi-Class Plant Disease Recognition and Classification for Diverse Crops

  • 期刊论文

  • 发表日期：

    2025年8月1日

查看全文
  • Sangamesh M Magi,
  • Jayashri Madalgi,
  • Lokesh. B Bhajantri

摘要

Plant diseases significantly impact agricultural productivity, necessitating timely and accurate diagnosis. While existing deep learning-based systems offer promising solutions, they often lack robustness, struggle with unseen crop types and diseases, and are generally limited to single-disease detection using leaf imagery alone. This study proposes a comprehensive multi-class plant disease recognition framework to address these limitations. A generalized deep learning model based on the Small Inception architecture was developed to classify healthy and diseased leaves across diverse crops by focusing on localized lesion regions. The approach utilizes a patch-based analysis strategy to detect multiple diseases on a single leaf while removing crop-specific dependencies. Experimental results on the PlantVillage dataset demonstrate superior performance: the Small Inception model achieved a training accuracy of 91.81% with a loss of 0.2385, and a testing accuracy of 90.57% with a loss of 0.2954. In comparison, MiniVGGNet and LeNet5 achieved lower testing accuracies of 80.24% and 80.72%, respectively. These results affirm the robustness and generalizability of the proposed system, which supports accurate multi-disease detection and extends applicability to diverse crop types and real-world farming conditions.