Few-Shot Object Detection for High-Speed Rail Infrastructure Defects

Ensuring the safety of high-speed rail infrastructure relies on effective defect detection. With the advancements in artificial intelligence, deep learning-based detection methods have been increasingly applied for daily inspections of high-speed rail infrastructure. However, detecting defects in high-speed rail infrastructure poses challenges due to limited samples, small targets, and difficulty in detection. To address these issues, this article proposes a few-shot object detection method based on a multi-scale attention mechanism, aiming to enhance the performance of defect detection in high-speed rail infrastructure. While most existing few-shot object detection methods employ two-stage models that require training on a large number of base class samples followed by fine-tuning on a small number of novel samples, we introduce gradient transmission separately for the gradient decoupled layer in Decoupled Faster R-CNN to better detect novel classes. Additionally, we incorporate the feature pyramid structure and Coordinate Attention into the original network architecture to enhance the detection of HSR infrastructure defects across different scales and improve the model's sensitivity to channel and location information. Experimental evaluations using UAV-captured image datasets, with normal class representing the base class and defects as the novel class, demonstrate the effectiveness of our proposed method.