Improving the method of straddle monorail PC track beam surface crack detection based on YOLOX
PC rail beam is an important part of the straddle monorail transit system,and its surface cracks directly threaten the safety of the entire traffic line.To address the problems of low accuracy and poor efficiency in traditional PC track beam surface crack detection,a track beam surface crack detection method based on an improved YOLOX was proposed.In the method,Ghost lightweight convolution was applied to extract target features in the enhanced feature extraction network,thereby reducing the model's computational complexity.The adaptively spatial feature fusion(ASFF)module was embedded in the feature fusion layer,allowing the model to adaptively learn the relationships between each feature layer,and the network paid more attention to the crack regions.Bicubic interpolation method was used for up-sampling to further improve the performance of network detection.The results of the experiment on the self-made track beam surface crack dataset show that compared with the original one,the average detection accuracy of the improved model increased by 4.40%,while the number of parameters decreased by 25.08%,and 119 frames of crack images were detected per second,indicating that the model meets the accuracy and real-time demands of practical engineering.
PC track beamcrack defect detectionadaptively spatial feature fusion(ASFF)bicubic interpolation method
NETL
NSTL
万方数据
