Research on Lightweight of Steel Surface Defect Detection Network Based on Improved YOLOv5
This research designs an improved lightweight network based on YOLOv5 model,which can quickly and accurately detect steel surface defects.Firstly,the ELAN structure based on gradient path design is used to improve the detection accuracy by improving the learning ability of the network;Secondly,the depth separable convolution and Ghostv2 module are introduced to reduce the volume and parameters of the model;Finally,the SIOU boundary box loss function is used to train the model,so that the model can quick-ly converge and accurately regress.The experimental results on NEU-DET show that the mAP value of the improved model is increased to 77.0%,which is 5.3%higher than the original model,the model volume is reduced by 42.1%,the number of parameters is reduced by 43.4%,and the detection speed is also 0.4 ms faster,realizing the balance between the lightweight effect of the model and the detection accuracy,and pro-viding a feasible scheme for subsequent deployment on the hardware terminal.
NETL
NSTL
万方数据
