Researchers have developed various models for detecting steel defects, including those that utilize multi-scale feature fusion, convolutional neural networks, and transformer dual-encoder models. These models, such as YOLO-MFD, MFDH-Net, and DEENet, aim to improve the accuracy and efficiency of steel surface defect detection. The specifics of each model differ, with some incorporating techniques like progressive dilation, anchor-free algorithms, and dynamic fine-grained multi-branch encoders.