| 引用本文: | 侯宪庆,黎远松,石 睿,王 涛.基于改进 YOLOv8s 的铝型材表面缺陷检测(J/M/D/N,J:杂志，M：书，D：论文，N：报纸).期刊名称,2026，43（2）：116-124 |
| CHEN X. Adap tive slidingmode contr ol for discrete2ti me multi2inputmulti2 out put systems[ J ]. Aut omatica, 2006, 42(6): 4272-435 |
|
| 摘要: |
| 目的 针对铝型材表面缺陷类别多,缺陷之间差异较大,容易出现漏检、误检等情况,提出一种改进的
YOLOv8s 检测算法。 方法 首先,引入扩张式残差模块 DWR(Dilation-wise Residual),基于此重新构造主干网络部
分 C2f 中的 Bottleneck 结构,增强网络对复杂特征的提取能力,使网络更高效地获取上下文信息;其次,在原颈部网
络中加入小目标检测层,旨在提取和传递那些在小尺寸缺陷中更为关键且判别性更强的小目标特征信息,将浅层
和深层特征进行融合,降低漏检率和误检率,提高检测精度; 最后,引入 Inner-GioU( Inner Generalized Intersection
over Union) 损失函数,关注边界框内部的重叠部分,加快模型收敛速度的同时提高边界框回归的准确性。 结果 将
改进算法应用在天池铝型材数据集中,实验结果显示:改进的算法在精确率、召回率和 mAP@ 0. 5 方面分别达到了
88. 7%、83. 4%和 88. 5%的性能指标,相比原始的 YOLOv8s 算法分别提高了 4. 1%、2. 1%和 2. 9%,改进算法能有效
识别铝型材表面不同类型的缺陷。 结论 通过实验可以证明改进算法的有效性,改善了铝型材部分缺陷检测效果较
差的问题,同时减少了漏检和误检情况,满足当前工厂对铝型材表面缺陷的检测要求。 |
| 关键词: 缺陷检测 YOLOv8s 深度学习 Inner-GIoU |
| DOI： |
| 分类号: |
| 基金项目: |
|
| Surface Defect Detection of Aluminum Profiles Based on Improved YOLOv8s |
|
HOU Xianqing LI Yuansong SHI Rui WANG Tao
|
|
School of Computer Science and Engineering Sichuan University of Science & Engineering Yibin 644000 Sichuan
China
|
| Abstract: |
| Objective Aiming at the problems that there are many types of surface defects on aluminum profiles and
significant differences among these defects which may easily lead to missed detections and false detections an improved
YOLOv8s detection algorithm is proposed. Methods First the dilation-wise residual DWR module was introduced.
Based on this the Bottleneck structure in the C2f of the backbone network was reconstructed to enhance the network?? s
ability to extract complex features and enable the network to obtain context information more efficiently. Second an
additional layer for detecting small targets was integrated into the existing neck network which aimed to extract and
transmit the small target feature information that was more critical and discriminative in small-sized defects. The shallow
and deep features were fused to reduce the missed detection rate and false detection rate and improve the detection
accuracy. Finally the inner generalized intersection over union Inner-GIoU loss function was introduced to focus on the overlapping part inside the bounding box accelerating the model?? s convergence speed while improving the accuracy of
bounding box regression. Results The improved algorithm was applied to the Tianchi aluminum profile dataset. The
experimental results showed that the performance indicators of the improved algorithm in terms of precision recall and
mAP@ 0. 5 reached 88. 7% 83. 4% and 88. 5% respectively. Compared with the original YOLOv8s algorithm these
values were increased by 4. 1% 2. 1% and 2. 9% respectively. The improved algorithm can effectively identify different
types of defects on the surface of aluminum profiles. Conclusion The effectiveness of the improved algorithm is verified by
experiments. This improved algorithm solves the problem of poor detection results for some aluminum profile defects
reduces missed and false detections and meets the current requirements of factories for detecting surface defects on
aluminum profiles. |
| Key words: defect detection YOLOv8s deep learning Inner-GIoU |
