| 摘 要: 针对YOLOv8在边缘特征提取方面的局限性,提出了一种改进的Backbone结构。设计了一种边缘特征增强模块,利用Sobel算子计算图像亮度梯度以识别边缘位置,并将边缘信息与纹理信息融合以增强图像特征的表示;引入了双层路由注意力机制,增强模型对关键信息的捕捉能力;对快速空间金字塔池化(Spatial Pyramid Pooling-Fast,SPPF)模块集成大核分离卷积注意力机制 (Large Kernel Separation Convolution Attention mechanism,LSKA),提升了模型对多尺度特征的聚合能力。实验结果表明,改进后模型精确率和平均精度均值分别提高了5.9%和1.9%,为钢材表面缺陷的检测任务提供了参考。 |
| 关键词: 缺陷检测 YOLOv8n 边缘特征增强 大核分离卷积 注意力机制 |
|
中图分类号:
文献标识码: A
|
| 基金项目: 河北省自然科学基金项目(2024209149);河北省省级科技计划资助项目(No6Z1019G) |
|
| Edge Feature Enhancement-Based Steel Surface Defect Detection Method |
|
LI Jiacheng, LONG Haiyang, GUI Yongliang, ZHANG Haodong
|
(School of Metallurgy and Energy, North China University of Science and Technology, Tangshan 063210, China)
15527170926@163.com; lhy@ncst.edu.cn; gyl@ncst.edu.cn; 485670285@qq.com
|
| Abstract: To address the limitations of YOLOv8 in edge feature extraction, an improved Backbone structure is proposed. Firstly, an edge feature enhancement module is designed, which utilizes the Sobel operator to compute image brightness gradients for identifying edge locations and fuses edge information with texture features to enhance the representation of image characteristics. Secondly, a dua-l layer routing attention mechanism is introduced to strengthen the model’s ability to capture critical information. Finally, a Large Kernel Separation convolution Attention mechanism (LSKA) is integrated into the Spatial Pyramid Pooling-Fast(SPPF)module, enhancing the model’s capability to aggregate mult-i scale features. Experimental results demonstrate that the improved model achieves increases of 5.9% in precision and 1.9% in mean Average Precision (mAP), providing a valuable reference for steel surface defect detection tasks. |
| Keywords: defect detection YOLOv8n edge feature enhancement large kernel separation convolution attention mechanism |
