Improved Oil Storage Tank Surface Defect Detection Algorithm Based on YOLOv5s Network
Aiming at the problems of low detection accuracy and slow detection efficiency of current oil tank surface defect detec-tion algorithms in practical applications,an improved oil tank surface defect detection algorithm based on YOLOv5s was proposed.By in-tegrating the global attention mechanism(GAM)and C3 structure,the information diffusion was effectively reduced,and the global di-mension interaction features were amplified to improve the detection efficiency.Part of the convolutional layers of the backbone network was replaced by an improved lightweight network RepVGG to enhance the feature extraction ability of the backbone network.Finally,an adaptive spatial feature fusion(ASFF)mechanism was used to improve the scale invariance of features,so that the shallow and deep feature maps could be fused more reasonably.The experimental results show that the mean average accuracy(mAP)of the improved al-gorithm on the self extracted dataset of petroleum industry is 92.5％,which is 2.8％higher than the original YOLOv5s.At the same time,the detection accuracy reaches 89.8％,which is 4.9％higher than the original YOLOv5s,further meeting the needs of surface de-fect detection for petroleum storage tanks.
万方数据
维普
