基于机器识别的带钢表面缺陷检测研究

Study on Surface Defect Detection of Strip Steel Based on Machine Identification

针对传统带钢表面缺陷检测技术落后、效率不高及小目标识别能力不足等问题,提出一种改进的YOLOv5s-Tiny目标检测模型,在保持模型较小计算量的同时提升检测速度和识别精度。通过将主干网络GSP-Darknet53替换为轻量级GhostNet网络,减少模型参数的数量,提高推理速度。在主干网络加入CBAM注意力机制,通过通道注意力机制和空间注意力机制对特征信息进行融合增强,提高小目标检测精度,并将损失函数GIoU改进为EIoU,提高检测框定位能力。最后将改善后的训练模型格式转换后安装到手机安卓端验证优化的有效性。结果表明:在东北大学数据集中,改进后模型检测精度提高1.5%的同时,召回率提升了1.5%,参数量减少12.3%;安卓端检测速度约为120 ms,完成带钢缺陷的实时检测。

In order to solve the problems of backward traditional strip surface defect detection technology,low efficiency and insuffi-cient small target identification ability,an improved YOLOv5s-Tiny target detection model was proposed,which improved the detection speed and recognition accuracy while maintaining the small calculation amount of the model.The backbone network GSP-Darknet53 was replaced with the lightweight GhostNet network to reduce the number of model parameters and improve the reasoning speed.The CBAM attention mechanism was added to the backbone network to enhance the feature information through channel attention mechanism and spatial attention mechanism to improve the detection accuracy of small target,and the loss function GIoU was improved to EIoU to im-prove the positioning ability of the detection box.Finally,the improved training model format was converted and installed to the Android terminal.The results show that in the Northeastern University data set,the mAP of the improved model is increased by 1.5%,the param-eter volume is reduced by 12.3%,the recall rate is increased by 1.5%,and the Android end detection speed is about 120 ms,which completes the real-time detection of strip steel defects.