Surface Defect Detection of Weldment Based on Improved YOLOv5
In order to improve the detection accuracy and efficiency of weldment surface defects in indus-trial environment,an improved YOLOv5 target detection model for weldment surface defect detection algo-rithm is proposed.Firstly,the C3 module in the central network is improved,including the introduction of ConvMixer,a mixed convolution structure,and Mish activation function,the Shuffle Attention is also added to the module.The improvement achieves better detection accuracy while reducing the complexity of the model.Besides,in order to overcome the shortcomings of NWD Loss,modification is made to focus more on the geometric information of bounding boxes.Lastly,all the standard convolutional layers in Neck are replaced with GSConv convolutional layers,which further enhances the network performance.Experimental results show that the mean average precision of the improved model reaches 91.3%,which is 4.8%higher than the original network.Besides,the amount of parameters and calculations of the model decrease by 21.4%and 8.9%respectively.And it has a detection frame rate of 142.9 f/s.The improved model not on-ly increases the detection accuracy,but also reduces the structural complexity,which meet the requirements for the detection of weldment surface defects in industrial production.