| 摘 要: 针对工业场景中钢材缺陷检测精度以及检测效率都很低的问题,提出了一种基于 YOLOv8n改进的工业缺陷检测算法。改进算法通过在主干网络末端嵌入卷积注意力融合模块(CAFM),实现全局与局部特征的有效融合,增强网络捕捉缺陷细节特征的能力。设计基于小波变换卷积(WTConv)的C2fWT模块,以增强网络对缺陷目标的多尺度特征提取能力。提出Shape-IoU损失函数以更加关注缺陷目标的形状与尺寸因素,改善缺陷边界框的回归效果。实验结果表明,与基准模型相比,改进算法平均精度提升了3.3%,计算量降低了6.2%,在精度与效率之间取得了较好平衡,具有重要的工业应用价值。 |
| 关键词: 缺陷检测 卷积注意力 小波变换 多尺度特征 |
|
中图分类号:
文献标识码: A
|
| 基金项目: 国家自然科学基金资助项目(52279069) |
|
| Industrial Steel Defect Detection Algorithm Based on Multi-Scale Features |
|
HUANG Zhiyong, LI Xiang, ZHANG Xu
|
(College of Computer and Information Technology, China Three Gorges University, Yichang 443002, China)
hzy@hzy.org.cn; 1627242777@qq.com; 649359073@qq.com
|
| Abstract: To address issues such as low accuracy and inefficiency in steel defect detection in industrial scenarios,an improved industrial defect detection algorithm based on YOLOv8n is proposed. The improved algorithm embeds a Convolutional Attention Fusion Module (CAFM) at the end of the backbone network to achieve effective fusion of global and local features, enhancing the network’s ability to capture detailed defect characteristics. A C2fWT module based on Wavelet Transform Convolution (WTConv) is designed to strengthen the network’s Mult-i Scale Feature extraction capability for defect targets. Additionally, the Shape-IoU loss function is introduced to place greater emphasis on the shape and size factors of defect targets, thereby improving the regression performance of defect bounding boxes. Experimental results demonstrate that, compared to the baseline model, the improved algorithm achieves a 3.3% increase in average precision and a 6.2% reduction in computational load, striking a favorable balance between efficiency and accuracy. This advancement holds significant industrial application value. |
| Keywords: defect detection convolutional attention wavelet transform Multi-Scale Features |
