Recognition Algorithm of Road Pavement Disease in Foggy Scene
Aiming at the problems of over-fitting,poor robustness and low accuracy of the target detection model based on deep learning throughout the process of road disease detection in foggy scene,a foggy road disease detection algorithm based on AECR-YOLOv5 fusion mechanism is proposed.Firstly,according to the characteristics of road pavement diseases,a data set expan-sion method depend on disease category is designed,and the expanded training set is obtained by this method.Meanwhile,clear picture and the depth picture after bilateral filtering are synthesized on the basis of atmospheric scattering model,so as to enhance the credibility of the synthetic foggy picture.Then,according to the uncertainty of task dependence,the loss functions of defogging network and detection network are weighted and fused to obtain the coupling loss function and promote the effective fusion of them.Finally,experiments are carried out by changing the input and network structure.Results show that when the input is the extended foggy picture training set and the network structure is AECR-YOLOv5 fusion model,its mAP on the test set is higher than other combination methods.
国家科技期刊平台
NSTL
万方数据
