Surface Object Detection Algorithm of Lightweight Steel Rolling Based on Machine Vision

Aiming at the problem of a large number of parameters and large computation in deep learning for steel rolling defect detection, a lightweight object detection algorithm was proposed based on YOLOv5s improved algorithm. Firstly, the YOLOv5s trunk feature extraction network was reconstructed with a simplified Efficientnet network. Then, coordinate attention (CA) was introduced to embed in the network to highlight the useful channel information. The improved Stem structure was introduced to replace the Focus structure to enhance the shallow feature extraction ability of the network. Finally, the final prediction box was generated with the decoder of YOLOv5. The experimental results showed that compared with YOLOv5s, the number of parameters and calculation amount of the improved YOLOv5s model were reduced by 53.0% and 57.0%, respectively, the mAP (mean average precision) was up to 80.2%, and the inference time was up to 30 ms frame−1. The experimental results showed that the improved model can be used to detect steel rolling defects in production, and it was possible to deploy the algorithm to handheld object detection devices with limited computing capacity and memory.