Saliency-Guided Deep Learning for Bridge Defect Detection in Drone Imagery

TL;DR

This paper tackles automatic detection, localization, and classification of concrete bridge defects in drone imagery using a two-stage, saliency-guided pipeline. The first stage computes local saliency maps by combining SmoothGrad with a linearity map to generate bounding-box proposals and creates saliency-enhanced images via bounding-box brightness augmentation, while the second stage applies a multi-label YOLOX detector to these enhanced images. Empirical results on CODEBRIM show that the saliency-guided approach outperforms baselines, achieving state-of-the-art mAP at IoU thresholds and an 8.3% improvement over YOLOX alone. The method holds practical potential for real-time, self-powered UAV inspections, reducing missed defects and enabling robust bridge maintenance workflows.

Abstract

Anomaly object detection and classification are one of the main challenging tasks in computer vision and pattern recognition. In this paper, we propose a new method to automatically detect, localize and classify defects in concrete bridge structures using drone imagery. This framework is constituted of two main stages. The first stage uses saliency for defect region proposals where defects often exhibit local discontinuities in the normal surface patterns with regard to their surrounding. The second stage employs a YOLOX-based deep learning detector that operates on saliency-enhanced images obtained by applying bounding-box level brightness augmentation to salient defect regions. Experimental results on standard datasets confirm the performance of our framework and its suitability in terms of accuracy and computational efficiency, which give a huge potential to be implemented in a self-powered inspection system.

Figures (4)

• Figure 1: Our proposed method. (a): Saliency and region proposal module; (b): Deep learning defect detection module.
• Figure :
• Figure :
• Figure :