Research On High - Resolution Remote Sensing Image Road Extraction Method

As one of important data information of the remote sensing image, the road has been the research hotspots and difficulty on image processing fields. The urban traffic management、urban planning vehicle automatic navigation^ and the updating of geographic information system database, etc all depend on the high precision automatic recognition and extraction of the roads and the timely updating of the road network information. Because of the technology limitation of computer> target detection and artificial intelligence, etc at present, there is no a set of perfect system can extract the road information intelligence and automated. It has a great significance to study the high resolution remote sensing image road extraction algorithm.The high resolution remote sensing image were used as the data source, and the algorithm of road information extraction and detection were studied and improved carefully, three different kinds of method were used to extract and detect the road in this article. Main research works were carried out as follows:(1) Achieve the road extraction with mathematical morphology and the improved Hough transform. The morphology was used on the image segmentation、 skeleton extraction and the optimization of late period, etc. The traditional Hough transform had been improved to extract the road information combing with morphology, the processing efficiency was raised, and obtained better effect at the same time.(2) FLD (Fisher Linear Discriminant) and using shape feature recognition to achieve road extraction. FLD was used to classify the images based on their color features to realize the road extraction with shape recognition and morphology optimization. To solve the threshold was difficult to determine on threshold segmentation in previous studies, a threshold calculation method was put forward, experiments showed, the effect was better to calculate the threshold segmentation with this method.(3) To extract the road by the improved SVM and FCM clustering algorithm. In order to solve the traditional problem that FCM was sensitive to the "different object with the same spectra characteristics" in images, pixel neighborhood spatial information was introduced, and a space function was defined to modify the FCM clustering. As a result, the effect of FCM clustering segmentation was improved, the combine of the improved FCM and SCM set the advantages of both supervised clustering and unsupervised clustering, which could realize the road extraction better.(4) Given experimental confirmation of the above three methods, and each method was tested with different images, the feasibility of the methods above were confirmed and the results of the experiments were analyzed.