Because of low detection accuracy, missing detection and false detection of strip steel surface defects, a new strip surface defect detection algorithm Gold-YOLOv8 is proposed. Firstly, the context anchor attention mechanism is combined with RepNCSPELAN4 module in YOLOv9 to replace C2f module in YOLOv8 to enhance the feature extraction capability. Secondly, the feature fusion of YOLOv8 is replaced by the information gather-and-distribute mechanism proposed by Gold-YOLO to reduce the information loss in the process of feature fusion. Finally, considering the quality unbalance of the defect samples, the Wise-IoU loss function is used, and its stepwise gain allocation strategy can solve the problem partly, and improve the convergence speed and regression accuracy. Compared with YOLOv8 on the NEU-DET dataset, Gold-YOLOv8 achieves an average detection accuracy of 82.7% and detection speed of 123.5 frames/s, which increases the average detection accuracy by about 5 percentage points and the detection speed by 8 times. The result shows that Gold-YOLOv8 can effectively improve the accuracy and speed of strip surface defect detection compared with other algorithms.