| 摘 要: 针对人造板连续压机生产线的钢带表面缺陷类别多、特征不明显导致的误检及漏检等问题,提出一种改进YOLOv8模型。引入扩张残差(Dilation-Wise Residual,DWR)模块,利用两步残差特征提取方法,提高分割中多尺度信息捕获效率;利用扩张残差分割(Dilation-Wise Residual Segmentation,DWRSeg)网络改进卷积模块,重构上下文特征融合网络的Bottleneck结构,提高网络对复杂特征的提取能力;使用强语义信息对特征上采样,拼接包含详细信息的特征,增强网络对上下文信息的获取效率。实验结果表明,6种钢带表面缺陷的平均精度均值mAP达到80.3%,与原YOLOv8模型的77.0%相比有一定增长。 |
| 关键词: 钢带 缺陷检测 目标检测 改进YOLOv8 卷积模块 注意力机制 |
|
中图分类号:
文献标识码: A
|
| 基金项目: 山东省产品质量检验研究院院储备项目(2024ZJKY0008);山东省产品质量检验研究院“建材家居产品检测中心科研创新团队”项目(20240901) |
|
| Steel Strip Surface Defect Detection Based on Improved YOLOv8 Model |
|
WANG Hua1, YANG Shengkun1, CAO Jun1, YANG Xiaoxia2, FANG Shuyu3, CAO Xue2, LI Xiang2
|
(1.Shandong Institute for Product Quality Inspection, Ji’nan 250102, China;
2.School of New Generation Information Technology, Shandong Polytechnic, Ji’nan 250104, China;
3.School of Information and Electrical Engineering, Shandong Jianzhu University, Ji’nan 250101, China)
wanghua@sdqi.com.cn; 18653118151@163.com; 15905317561@163.com; yangxiaoxia8899@126.com; fangshuyu_2024@163.com; cxctt678@163.com; lx24642024@163.com
|
| Abstract: To address issues such as false detection and missed detection caused by the numerous categories and inconspicuous features of steel strip surface defects in continuous press production lines for artificial boards, an improved YOLOv8 model is proposed. The Dilation-Wise Residual (DWR) module is introduced, leveraging a two-step residual feature extraction method to enhance the efficiency of mult-i scale information capture during segmentation. The DWRSeg network is utilized to refine the convolutional module, reconstructing the Bottleneck structure of the context feature fusion network to improve the model’s ability to extract complex features. Strong semantic information is employed for feature upsampling,which is then concatenated with detai-l rich features to boost the network’s efficiency in acquiring contextual information.Experimental results demonstrate that the mean average precision (mAP) for six types of steel strip surface defects reaches 80.3%, showing a notable increase compared to the original YOLOv8 model’s 77.0% . |
| Keywords: steel strip defect detection object detection improved YOLOv8 convolutional module attention mechanism |
