Scholarlink

<p>To enhance the accuracy and efficiency of image recognition for dense Agaricus bisporus, this study proposes an improved YOLOv8n model. First, a dataset containing a large number of Agaricus bisporus images at various positions is constructed. In terms of model improvement: (1) The detection head is optimized by introducing LADH (Lightweight Attention Detection Head) to improve the recognition capability of small objects, replacing ordinary convolutions with depthwise separable convolution (DSConv) to reduce computational load, while deepening the regression branch and simplifying the classification branch; (2) The backbone network is lightweighted by integrating Ghost modules with C3 modules, and substituting standard convolutions with Ghost convolutions to compress parameters and FLOPs . Experiments show that the improved YOLOv8n model achieves a mAP@0.5 of 91.1% in the recognition task of dense Agaricus bisporus, with 43.3% fewer parameters, 38.3% less FLOPs, 38.7% smaller model size, and 0.5 milliseconds shorter inference time compared to the original algorithm. This research provides effective support for intelligent picking in the industrialized cultivation of Agaricus bisporus.</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

Research on Agaricus Bisporus Image Recognition Based on Improved YOLOv8n Model

  • 期刊论文

  • 发表日期：

    2025年10月3日

查看全文
  • Jiawei Qin,
  • Chunyan Zhang,
  • Huanhuan Zhang

摘要

To enhance the accuracy and efficiency of image recognition for dense Agaricus bisporus, this study proposes an improved YOLOv8n model. First, a dataset containing a large number of Agaricus bisporus images at various positions is constructed. In terms of model improvement: (1) The detection head is optimized by introducing LADH (Lightweight Attention Detection Head) to improve the recognition capability of small objects, replacing ordinary convolutions with depthwise separable convolution (DSConv) to reduce computational load, while deepening the regression branch and simplifying the classification branch; (2) The backbone network is lightweighted by integrating Ghost modules with C3 modules, and substituting standard convolutions with Ghost convolutions to compress parameters and FLOPs . Experiments show that the improved YOLOv8n model achieves a mAP@0.5 of 91.1% in the recognition task of dense Agaricus bisporus, with 43.3% fewer parameters, 38.3% less FLOPs, 38.7% smaller model size, and 0.5 milliseconds shorter inference time compared to the original algorithm. This research provides effective support for intelligent picking in the industrialized cultivation of Agaricus bisporus.