Lightweight Pyramid Cross-Attention Network for No-Service Rail Surface Defect Detection
Vision-based rail defect detection plays a crucial role in ensuring the safety and efficiency of railway transportation systems. However, many existing methods face challenges such as high parameters, complex computation, slow inspection speed, and low accuracy. To tackle these challenges, this pape...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11075659/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Vision-based rail defect detection plays a crucial role in ensuring the safety and efficiency of railway transportation systems. However, many existing methods face challenges such as high parameters, complex computation, slow inspection speed, and low accuracy. To tackle these challenges, this paper introduces the Lightweight Pyramid Cross-Attention Network (LPCANet) for rail defect detection using RGB and depth images. Initially, a Lightweight Pyramid Module (LPM) is designed to extract multi-scale feature maps from depth images, while a backbone is used to capture pyramid feature details from RGB images. Subsequently, the cross-attention module (CAM) is applied across the two types of feature maps. Secondly, a structural feature extractor (SFE) is required to improve detection performance. Finally, a pixel shuffle operation is applied to restore the size of maps to align with the size of the ground truth labels (GT). Experimental results show that the LPCANet surpasses the performance of 18 state-of-the-art methods, with parameters of 9.90 M and a running speed of 162.60 frames per second (FPS). |
|---|---|
| ISSN: | 2169-3536 |