A Lightweight Model for Detecting Overlapping Anomalies in Steel Sections Based on YOLOv5

Aiming at the problems of overlapping anomalies of steel profiles, complex model structure and large number of model parameters in the current production process of steel profiles, an improved lightweight target detection model of yolov5s is proposed. Firstly, the MobileNetV3 network is used to replace the backbone network of yolov5s, which drastically reduces the number of parameters in the model and the model size; secondly, the C3 model is replaced by the C2f model to improve the accuracy of detection; finally, the CA attention mechanism is introduced to enhance the detection capability of small-scale targets, which further improves the accuracy. The accuracy of the improved model can reach 97.2%, meanwhile, the amount of parameters is reduced by 76.8%, the model size is reduced by 75%, and the GFLOPs are reduced by 81.6%. The experiments show that the improved yolov5s model is lightweight while maintaining high accuracy and is easy to deploy, which meets the requirements of real production.