Research on Detection Algorithm of Steel Surface Defects Based on GEC-YOLO Network Model

Due to the imperfect manufacturing process and external factors, there may be some defects on the steel surface, which seriously affects the service life and availability of steel. Correct and rapid detection of steel surface defects can greatly improve productivity and product quality. In this paper, an improved GEC-YOLO model is proposed. Firstly, a GAM module is integrated into the backbone feature extraction network to dynamically adjust the feature response and focus on the important parts of the image more accurately. Secondly, in the feature extraction network, EMA module is introduced to encode global information, recalibrate the weights of each channel, and help the model enhance the ability of feature expression. Finally, the light downsampling operator CARAFE is replaced by the original downsampling operator to reduce the loss of image information during upsampling. The test results show that the proposed model achieves 82.0% mAP on the NEU-DET dataset, which is 4.2% higher than that of the original network, and the FPS reaches 86.95 frames per second, which improves the detection ability of various types of steel strip defects and can meet the needs of real-time detection.