To investigate the safety of railway operations, it is significant to measure the technical defects of the rail transit. Traditional machine vision technology does not show better detection results due to the complicated and distinct features of the rail surface. This paper presents a deep learning algorithm based on YOLOv4 which better realise the rail surface's defect detection. Our proposed approach uses modified version of CSPDarkent-53 as the backbone, consequently achieve faster and more accurate detection outcomes. We use six types of datasets of different images of the rail surface for investigation of the defects and compare the outcomes with a different network. This technique is used to evaluate our proposed method's efficiency to detect and classify the defects by comparing them with other techniques. Moreover, the experimental outcomes depict the promising capability of our proposed method. The results represent that accuracy of this algorithm can reach up to 96.5% on the mAP metric. This approach has significant importance for the surface detection of defects in rail transit.
YOLOv4 Based Deep Learning Algorithm for Defects Detection and Classification of Rail Surfaces
2021-09-19
1038995 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Pedestrian Detection Algorithm Based on Improved YOLOv4
IEEE | 2023
|UAV Target Detection Algorithm Based on Improved YOLOv4
Springer Verlag | 2022
|