Abstract
Combining an object-detection algorithm with an unmanned aerial vehicle (UAV) can accelerate the detection of road cracks. To address the difficulties of intricate crack morphology, similar color to the road, and small crack area, this paper describes a UAV road crack object-detection algorithm using MUENet. The MUENet is primarily comprised of a main and auxiliary dual-path module (MADPM), an uneven fusion structure with transpose and inception convolutions (TI-UFS) and a E-SimOTA strategy. First, the MADPM is proposed to efficiently extract the essential morphological features of cracks. Subsequently, the TI-UFS is proposed to explore potential crack color characteristics. Finally, the E-SimOTA strategy accurately differentiates different types of cracks and accelerates network training convergence. The experimental results demonstrate that MUENet has the double benefits of precision and speed on a self-built dataset of UAV near-far scene images (UNFSI). This object-detection algorithm is more adaptable to crack objects than other mainstream object-detection algorithms.
•MADPM is proposed to address the problem of complicated crack morphology.•TI-UFS is proposed to tackle the issue of cracks resembling road colors.•E-SimOTA is proposed to address the problem of small crack area.