Scholarlink

<p>Potato scab is an increasingly prevalent disease that threatens tuber quality and market value. Although RGB imaging captures surface texture effectively, its contrast for scab lesions is often limited, particularly for small symptoms. Narrow-band imaging at 635&#xa0;nm provides complementary advantages by highlighting lesion-related biochemical changes. Leveraging the strengths of both modalities, this study develops an integrated system for automated scab detection and severity assessment. A dual-branch fusion network is designed to combine RGB and 635&#xa0;nm images at the feature level, enhancing lesion representation in the improved lightweight YOLOv8 architecture. The model is trained on a self-constructed potato scab dataset and achieves an mAP of 81.9%. Experimental results show a 3.8% increase in recall and a 1.5% gain in mAP50 over the original model, while reducing parameter size to 38%. Compared with other detection models, recall and mAP50 improve by 6.1%–7.9% and 1.4%–4.2%, respectively. Beyond detection, this study establishes a quantitative framework for disease severity evaluation. A comprehensive index, PSCSI, is proposed to jointly consider scab count, lesion size, and spatial distribution. Objective weighting is applied after parameter preprocessing, and spatial distribution is quantified using Density Peaks Clustering. The resulting index achieves more consistent and reliable grading than raw individual parameters. Overall, this work presents a complete pipeline for potato scab monitoring that integrates multimodal imaging, lightweight detection, and quantitative severity assessment. The approach provides a practical foundation for automated grading and disease management in potatoes and other agricultural products.</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 integrated imaging and deep learning system for automated detection and severity analysis of potato scab

  • 期刊论文

  • 发表日期：

    2026年5月5日

查看全文
  • Fengnong Chen,
  • Bo Wu,
  • Pengcheng Wu,
  • Zhengtao Feng,
  • Kai Ren,
  • Ye Li,
  • Hongwei Sun

摘要

Potato scab is an increasingly prevalent disease that threatens tuber quality and market value. Although RGB imaging captures surface texture effectively, its contrast for scab lesions is often limited, particularly for small symptoms. Narrow-band imaging at 635 nm provides complementary advantages by highlighting lesion-related biochemical changes. Leveraging the strengths of both modalities, this study develops an integrated system for automated scab detection and severity assessment. A dual-branch fusion network is designed to combine RGB and 635 nm images at the feature level, enhancing lesion representation in the improved lightweight YOLOv8 architecture. The model is trained on a self-constructed potato scab dataset and achieves an mAP of 81.9%. Experimental results show a 3.8% increase in recall and a 1.5% gain in mAP50 over the original model, while reducing parameter size to 38%. Compared with other detection models, recall and mAP50 improve by 6.1%–7.9% and 1.4%–4.2%, respectively. Beyond detection, this study establishes a quantitative framework for disease severity evaluation. A comprehensive index, PSCSI, is proposed to jointly consider scab count, lesion size, and spatial distribution. Objective weighting is applied after parameter preprocessing, and spatial distribution is quantified using Density Peaks Clustering. The resulting index achieves more consistent and reliable grading than raw individual parameters. Overall, this work presents a complete pipeline for potato scab monitoring that integrates multimodal imaging, lightweight detection, and quantitative severity assessment. The approach provides a practical foundation for automated grading and disease management in potatoes and other agricultural products.