Abstract: Disease assessment is an important aspect of roads and bridges maintenance. Current disease detection is mainly based on automatic measuring vehicles and visual judgments, with the shortcomings of heavy workload and high risk. Correspondingly, the low-flying six-rotor unmanned aerial vehicles (UAV) can take photos of roads and bridges from multi-angles, which has great advantages on roads and bridges detection. Based on UAV images, this paper developed a new method in disease detection of roads and bridges. First, the multi-component deformation model was used to simulate the disease target. Then the global image was searched to detect the potential disease areas. Finally, the disease areas were detected from UAV images. Experiments showed that the proposed algorithm could effectively detect the disease in complex background, and the target detection accuracy was over 80%, with high efficiency and strong robustness.

Key words: unmanned aerial vehicle images, disease detection, multi-component deformation model, feature pyramid