基于改进YOLOv5s的活塞杆表面缺陷检测

Surface defect detection of piston rod based on improved YOLOv5s

活塞杆是工业设备上技术含量较高的关键部件,但目前为止对其表面缺陷的检测还是以人工检测为主。针对人工检测方法准确性差、效率低下的情况,提出了一种基于改进YOLOv5s的活塞杆表面缺陷检测技术。首先,在Backbone部分引入SE和CBAM双注意力机制,其中CBAM注意力机制与C3模块相结合形成了新CBAMC3模块,提升算法对于缺陷信息的提取能力,从而进一步提高算法精度;其次,改善激活函数为GELU函数避免梯度消失,使算法有较好表现;最后,使用GSConv卷积模块代替Neck部分中的Conv卷积模块,降低计算成本,并且引入VoV-GSCSP模块,减少算法参数量,在轻量化算法的同时保持精度。试验结果表明,改进的YOLOv5-CSGGV算法总平均精度达到了81.1%,较原YOLOv5s算法提升了6.3%,算法参数量相比YOLOv5-SC算法减少了14.7%,使检测速度和精度达到了更好平衡,满足活塞杆工业生产过程中缺陷检测的要求。

The piston rod is a key component with high technical content in industrial equipment.However,so far,the detection of its surface defects is mainly manual inspection,and a piston rod surface defect detection technology based on improved YOLOv5s was proposed to solve the problem of poor accuracy and low efficiency of the manual inspection method.Firstly,the SE and CBAM dual attention mechanism was introduced in the Backbone part,in which the CBAM attention mechanism was combined with C3 module to form the new CBAMC3 module,which improves the algorithm′s ability to extract defect information,thereby further improving the algorithm accuracy.Secondly,the activation function was improved to the GELU function to avoid gradient disappearance,so that the algorithm has better performance.Finally,the GSConv convolution module was used to instead of the Conv convolution module in the Neck part,which reduces the calculation cost and the VoV-GSCSP module was introduced to reduce the number of parameters of the algorithm and lightweight the algorithm while maintaining accuracy.The experiment result shows that the average accuracy of the improved YOLOv5-CSGGV model reaches 81.1%,which is 6.3%higher than the original YOLOv5s algorithm,and the number of parameters of the algorithm decreases by 14.7%compared to the YOLOv5-SC algorithm,achieving a better balance between detection speed and accuracy,meeting the requirements for defect detection in the industrial production process of piston rods.