Steel Surface Defect Detection Method Based on MAA_YOLOv8

Abstract

In response to the problems of missed detection, high false detection rate, and difficulty in deploying models on edge devices in steel plate surface defect detection, this paper proposes a multi attention mechanism lightweight steel surface defect detection algorithm based on YOLOv8 (MAAYOLOv8). The model incorporates multiple attention mechanisms (including SimAM, SE, CBAM, ECA, and others) into both its backbone network and detection head. By leveraging GhostConv alongside an enhanced spatial pyramid pooling module, it not only boosts the capability of fine-grained feature extraction but also achieves a substantial reduction in parameter count. The experimental results showed that MAAYOLOv8 achieved a performance of 0.798 on the NEU-DET steel plate surface defect dataset mAP@0.5 Compared to YOLOv8n, it has increased by 4.7 percentage points, with an overall F1 score of 0.75. The model parameter count is only 2.1×106 and the computational complexity is 5.1×109, which are 34.4% and 41.4% lower than the original YOL0v8n, respectivel. A large number of visualized detection results and loss curves further validate the high robustness and practicality of the model in practical industrial scenarios, making it more suitable for deployment on edge devices and having important engineering application value.