摘要
在复杂工业场景下,传统目标检测算法对微小缺陷的识别精度不足,健壮性弱。基于此,提出一种基于改进Canny-YOLO算法的弹壳表面缺陷检测方法,融合Canny边缘检测与YOLOv5模型的多尺度特征提取优势。改进的Canny算法通过自适应阈值调整与多方向梯度响应优化,增强对裂纹、飞边等微小缺陷的边缘感知能力。YOLOv5模型引入多通道输入与注意力机制,实现语义特征与边缘特征的动态融合,并通过锚框优化提升小目标定位精度。实验表明,改进算法在复杂光照与表面反光场景下表现出优异的健壮性与实时性,为军工自动化生产线提供了高效可靠的检测方案。
In complex industrial scenarios,traditional object detection algorithms have insufficient recognition accuracy and weak robustness for small defects.Based on this,a shell surface defect detection method based on improved Canny-YOLO algorithm is proposed,which combines the advantages of Canny edge detection and YOLOv5 model for multi-scale feature extraction.The improved Canny algorithm enhances the edge perception ability of small defects such as cracks and flying edges through adaptive threshold adjustment and multi-directional gradient response optimization.The YOLOv5 model introduces multi-channel input and attention mechanism to achieve dynamic fusion of semantic features and edge features,and improves the accuracy of small target localization through anchor box optimization.Experiments have shown that the improved algorithm exhibits excellent robustness and real-time performance in complex lighting and surface reflection scenarios,providing an efficient and reliable detection solution for military automation production lines.
作者
罗海峰
LUO Haifeng(Guangdong Hongda Defense Technology Co.,Ltd.,Qingyuan 513325)
出处
《现代制造技术与装备》
2025年第9期82-84,共3页
Modern Manufacturing Technology and Equipment
