Scholarlink

<p>Apple leaf diseases seriously affect apple yield and fruit quality. Timely detection is therefore important for precision orchard management. In practical applications, apple leaf disease detection requires high accuracy, low parameter count, and low computational cost for mobile and edge deployment. However, existing lightweight redesigns often weaken feature representation, limiting detection performance. To address this issue, this paper proposes a&#xa0;lightweight network for apple leaf disease detection, termed Lightweight Multi-Scale Direction-Aware You Only Look Once (LMSD-YOLO). Built on YOLOv12n, LMSD-YOLO introduces Space-to-Depth Convolution (SPDConv) during downsampling to reduce fine-grained information loss. It designs a&#xa0;Multi-Scale Directional Attention Block (MSDA) and embeds it into the backbone and feature fusion paths to enhance multi-scale and directional feature representation. In addition, a&#xa0;lightweight detection head, termed LiteMBConv-Head, is designed to reduce the computational cost of the detection head while maintaining competitive localization accuracy. On the self-constructed P3&#xa0;Apple Leaf Disease (P3ALD) dataset, LMSD-YOLO achieved&#xa0;a&#xa0;mean average precision&#xa0;(mAP50) of 93.2% and an mAP50–95 of 63.9%. On the public Apple Leaf Disease Object Detection (ALDOD) dataset, it achieved an mAP50 of 95.6% and an mAP50–95 of 93.2%. The proposed model contains 1.39 M parameters and requires 4.0&#xa0;giga floating-point operations (GFLOPs). These results indicate that LMSD-YOLO maintains competitive detection performance under limited parameter and computational budgets, providing an efficient lightweight solution for apple leaf disease detection in resource-constrained scenarios.</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 Lightweight Deep Learning Network for Apple Leaf Disease Detection in Complex Orchard Scenes

  • 期刊论文

  • 发表日期：

    2026年7月1日

查看全文
  • Dongping Jiang,
  • Honghan Li,
  • Ji Zhao

摘要

Apple leaf diseases seriously affect apple yield and fruit quality. Timely detection is therefore important for precision orchard management. In practical applications, apple leaf disease detection requires high accuracy, low parameter count, and low computational cost for mobile and edge deployment. However, existing lightweight redesigns often weaken feature representation, limiting detection performance. To address this issue, this paper proposes a lightweight network for apple leaf disease detection, termed Lightweight Multi-Scale Direction-Aware You Only Look Once (LMSD-YOLO). Built on YOLOv12n, LMSD-YOLO introduces Space-to-Depth Convolution (SPDConv) during downsampling to reduce fine-grained information loss. It designs a Multi-Scale Directional Attention Block (MSDA) and embeds it into the backbone and feature fusion paths to enhance multi-scale and directional feature representation. In addition, a lightweight detection head, termed LiteMBConv-Head, is designed to reduce the computational cost of the detection head while maintaining competitive localization accuracy. On the self-constructed P3 Apple Leaf Disease (P3ALD) dataset, LMSD-YOLO achieved a mean average precision (mAP50) of 93.2% and an mAP50–95 of 63.9%. On the public Apple Leaf Disease Object Detection (ALDOD) dataset, it achieved an mAP50 of 95.6% and an mAP50–95 of 93.2%. The proposed model contains 1.39 M parameters and requires 4.0 giga floating-point operations (GFLOPs). These results indicate that LMSD-YOLO maintains competitive detection performance under limited parameter and computational budgets, providing an efficient lightweight solution for apple leaf disease detection in resource-constrained scenarios.