In the realm of railroad transportation, the switch sliding baseplate constitutes one of the most crucial components within railroad crossings. Wear defects occurring on the switch sliding baseplate can give rise to issues such as delayed switch operation, inflexible switching, or even complete failure, thereby escalating the risk of train derailment. Consequently, the detection of wear defects on the switch sliding baseplate is of paramount importance for enhancing traffic efficiency and guaranteeing the safety of train switching operations. Micro-cutting defects, which are among the most significant defects resulting from wear, exhibit complex and diverse morphological and characteristic features. Traditional random sampling methods struggle to capture their detailed characteristics, leading to inadequate accuracy and robustness in the detection process. To address the above-mentioned issues, the YOLOv5s algorithm has been refined and subsequently applied to the detection of micro-cutting defects generated by wear on the switch sliding baseplate. The experimental results demonstrate that, in comparison with the currently prevalent mainstream target detection algorithms, the improved model can attain optimal recall rates R, mAP@0.5, and mAP@0.5:0.95. Specifically, when contrasted with the original YOLOv5s algorithm, the improved model witnesses significant enhancements in its precision rate P, the recall rate R, mAP@0.5, and mAP@0.5:0.95, with increments of 1.26%, 5.6%, 9.1%, and 8.92%, respectively. These improvements fully corroborate the performance of the proposed model in the context of micro-cutting defect detection.