基于YOLOX的重载铁路钢轨伤损智能识别方法OA

To address the low efficiency of manual interpretation and the high miss rate of small-scale defects in ultrasonic rail inspection,this paper proposes an intelligent defect recognition method for rail ultrasonic B-scan images based on YOLOX.First,a multi-class dataset of rail ultrasonic B-scan images is constructed from field inspection data,and frequency-domain Gaussian low-pass filtering is applied to suppress high-frequency noise.Then,within the YOLOX framework,anchor-free bounding box regression,a decoupled detection head,and the SimOTA dynamic sample assignment strategy are adopted to improve the localization and classification stability of weak-reflection and small-scale defects.Comparative experiments with Faster R-CNN,RetinaNet,and SSD are conducted to systematically evaluate the performance of the proposed method in multi-class rail defect recognition.The results demonstrate that the proposed approach achieves superior performance in terms of accuracy,precision,recall,and F1-score,and exhibits more stable detection capability for small-scale defects such as rail head shelling,bolt-hole cracks,and rail foot micro-cracks.