Scholarlink

<p>Plant diseases represent a&#xa0;significant challenge to global agriculture, leading to substantial yield and quality losses if not detected early. This paper presents an end-to-end framework for plant leaf disease identification and severity grading from a&#xa0;single red–green–blue (RGB) image. The system integrates a&#xa0;lightweight You Only Look Once (YOLO) v8n classification model (YOLOv8n-cls) for simultaneous crop and disease recognition with a&#xa0;hybrid hue–saturation–value (HSV)–Lab color-space segmentation pipeline refined using GrabCut to isolate leaf and lesion regions. The percentage infected area (PIA) is computed and mapped into interpretable severity categories—Mild, Moderate, and Severe—using a&#xa0;fuzzy logic inference engine, while a&#xa0;consistency layer compares predicted labels with measured lesion ratios to reduce borderline misclassifications. The complete pipeline is implemented as a&#xa0;Streamlit-based application providing visual overlays, confidence scores, severity grades, and exportable reports. Experimental evaluation on the PlantVillage benchmark dataset achieved a&#xa0;Top‑1 classification accuracy of 98.7% on the held-out test split, demonstrating the feasibility and interpretability of the proposed approach under controlled dataset 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

YOLO-Based Plant Disease Identification with Fuzzy Logic Severity Grading

  • 期刊论文

  • 发表日期：

    2026年3月24日

查看全文
  • Ravula Navaneeth,
  • Chinta Sai Sri Sagar,
  • Pulicharla Nikhitha,
  • Saanvika Uddagiri,
  • S. Manimaran

摘要

Plant diseases represent a significant challenge to global agriculture, leading to substantial yield and quality losses if not detected early. This paper presents an end-to-end framework for plant leaf disease identification and severity grading from a single red–green–blue (RGB) image. The system integrates a lightweight You Only Look Once (YOLO) v8n classification model (YOLOv8n-cls) for simultaneous crop and disease recognition with a hybrid hue–saturation–value (HSV)–Lab color-space segmentation pipeline refined using GrabCut to isolate leaf and lesion regions. The percentage infected area (PIA) is computed and mapped into interpretable severity categories—Mild, Moderate, and Severe—using a fuzzy logic inference engine, while a consistency layer compares predicted labels with measured lesion ratios to reduce borderline misclassifications. The complete pipeline is implemented as a Streamlit-based application providing visual overlays, confidence scores, severity grades, and exportable reports. Experimental evaluation on the PlantVillage benchmark dataset achieved a Top‑1 classification accuracy of 98.7% on the held-out test split, demonstrating the feasibility and interpretability of the proposed approach under controlled dataset conditions.