A Novel UAV-Based Road Damage Detection Algorithm with Lightweight Convolution and Attention Mechanism

In this paper, a novel attention- and lightweight convolution-based road damage detection network (ALC-Net) is proposed to address the trade-off between accuracy and real-time performance in processing unmanned aerial vehicle (UAV) imagery. Specifically, a lightweight module that integrates ghost convolution with the squeeze-and-excitation (SE) attention mechanism is designed, which effectively reduces model parameters while enhancing detection accuracy. The focus module is introduced to perform downsampling and channel-wise concatenation of input images, thereby enriching feature diversity. Furthermore, a coordinate attention mechanism is incorporated to aggregate horizontal and vertical spatial information, emphasizing subtle road damage characteristics. The proposed ALC-Net is comprehensively evaluated on a UAV-captured road damage dataset, demonstrating superior detection performance compared to other state-of-the-art approaches. The contributions of key components in ALC-Net are also validated through ablation studies, confirming their ability to enhance feature extraction capabilities while reducing computational complexity. Additionally, experiments on non-UAV road damage datasets further reveal the robust generalization capability of ALC-Net, exhibiting substantial potential for broader applications.