Abstract: In order to solve the problem that current steel surface defect detection methods are ineffective in detecting small defects, an algorithm for detecting small defects on the steel surface that integrates Fourier convolution and difference perception is proposed. The algorithm uses CSP-FFCM to replace the BasicBlock in the backbone network, and performs convolution operations in the spatial and frequency domains to reduce the computational overhead and enhance the feature extraction capability of the network. Then, a multi-scale feature layer optimization strategy is proposed, which optimizes the allocation of computational resources while preserving fine-grained feature information to ensure that the model effectively captures the detailed information of tiny defects. Finally, a difference-aware feature enhancement module is designed to further improve the model’s detection performance of tiny defects by strengthening the feature representation capability of tiny defects. The experimental results show that the algorithm achieves mAP indexes of 83.7% and 73.1% on the NEU-DET and GC10-DET datasets, respectively, and exhibits significant performance advantages in the task of high-precision detection of tiny defects on steel surfaces.

Key words: minor defect detection, Fourier convolution, multi-scale feature layer optimisation, difference perception