Leveraging Saliency in Single-Stage Multi-Label Concrete Defect Detection Using Unmanned Aerial Vehicle Imagery

Visual inspection of concrete structures using Unmanned Areal Vehicle (UAV) imagery is a challenging task due to the variability defects’ size and appearance. This paper proposes high-performance model for automatic fast detection bridge defects UAV-acquired images. Our method, coined Saliency-based Multi-label Defect Detector (SMDD-Net), combines pyramidal feature extraction attention through one-stage defect model. The module extracts local global saliency features, which are scaled integrated with network max-pooling, multiplication, residual skip connections operations. has effect enhancing localisation small low-contrast defects, as well overall accuracy in varying image acquisition ranges. Finally, multi-label loss function used identify localise overlapping defects. experimental results on standard dataset real-world images demonstrated performance SMDD-Net regard state-of-the-art techniques. computational efficiency make it suitable method UAV-based structure inspection.