Scholarlink

<p>Lightweight edge detection has become a critical research focus in computer vision; however, it is often accompanied by a decline in feature extraction capabilities. To address this, we propose LBVNet (Bio-inspired Visual Mechanism Lightweight Network), a bio-inspired lightweight edge detection network. Inspired by the non-classical receptive field suppression and disinhibition mechanisms in V1 cells of the biological visual system, we design a positive–negative modulation factor convolutional module to enhance edge information extraction. Additionally, we introduce a feedback modulation feature integration module in the decoding network to improve multi-scale feature fusion efficiency. Experimental results show that LBVNet achieves an ODS F-measure of 0.802 on the BSDS500 dataset with only 0.25M parameters, and performs excellently on the NYUD and Multicue datasets. This network offers a new bio-visual heuristic solution for low-parameter edge detection, opening up new research avenues. Through the interdisciplinary integration of neurobiology and deep learning, we propose a lightweight edge detection solution that combines biological plausibility with computational efficiency.</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

Bio-inspired visual mechanism lightweight network for edge detection

  • 期刊论文

  • 发表日期：

    2025年4月2日

查看全文
  • Kaijian Huang,
  • Chuan Lin,
  • Jiansheng Peng

摘要

Lightweight edge detection has become a critical research focus in computer vision; however, it is often accompanied by a decline in feature extraction capabilities. To address this, we propose LBVNet (Bio-inspired Visual Mechanism Lightweight Network), a bio-inspired lightweight edge detection network. Inspired by the non-classical receptive field suppression and disinhibition mechanisms in V1 cells of the biological visual system, we design a positive–negative modulation factor convolutional module to enhance edge information extraction. Additionally, we introduce a feedback modulation feature integration module in the decoding network to improve multi-scale feature fusion efficiency. Experimental results show that LBVNet achieves an ODS F-measure of 0.802 on the BSDS500 dataset with only 0.25M parameters, and performs excellently on the NYUD and Multicue datasets. This network offers a new bio-visual heuristic solution for low-parameter edge detection, opening up new research avenues. Through the interdisciplinary integration of neurobiology and deep learning, we propose a lightweight edge detection solution that combines biological plausibility with computational efficiency.