A three-stage"preprocessing-inference based on neural network-post-processing"road extraction algorithm is pro-posed to deal with the existing problems of road extraction algorithms,such as road breakage,noise and other problems.Firstly,ac-cording to the characteristics of aerial images,image enhancement is used to maximally eliminate the impact of environmental fac-tors on the images.Then,a deep learning-based inference network is designed to extract road information from the preprocessed im-ages.Finally,a morphology-based post-processing algorithm is proposed to further correct the road information.The experimental results show that the proposed algorithm has higher performance compared to the existing algorithms.
万方数据