Study On Segmentation Algorithms And Its Applications Of High Spatial Resolution Remote Sensing Images

With the development of remote sensing technology,the spatial resolution of remote sensing images are more and more high.The high spatial resolution images bring opportunities for the development of remote sensing technology,while at the same time they also bring new challenges for the processing of remote sensing data because of their own characteristics,which cause the traditional pixel level processing methods can not adapt to the current demand of remote sensing analysis and applications.In this case,the object based technology has become a new choice for the analysis of high resolution remote sensing images.However,object oriented analysis is not mature in theory and technology,there are still many problems to be studied and solved,especially for the basal step of image segmentation.At present,there are still some limitations in different degree for the study of image segmentation,such as lack of segmentation accuracy and efficiency,poor adaptability of the model,lack of unified scale evaluation standards and so on.Aiming at the existing deficiencies in current research work,this paper mainly explores and discusses the problems of segmentation for high spatial remote sensing images and the choice of optimal segmentation scale in multi-scale segmentation methods,the main research contents and results are as follows:(1)A fast segmentation method for remote sensing images based on region merging is proposed.In this method,the initial segmented regions are obtained using the watershed transform,and then they are merged by the local optimization strategy to obtain the final segmentation results.In the process of region merging,the data structure of neighbor array is designed to maintain the adjacency relations between regions for the low efficiency of traditional region adjacency graph,and compared to adjacency graph,the data structure has higher efficiency.In order to imporve the accuracy of the algorithm,the region merging process is divided into two steps,and the spectral,geometry and texture information is synthetically concerned in the merging criterion.By comparing with the traditional segmentation algorithm based on adjacency graph,it is proved that the algorithm is superior in efficiency and accuracy(2)A multi scale segmentation method for fusing multiple image features is proposed.The method also belongs to the regional segmentation algorithm,in order to solve the low precision problem causing by insufficient use of image features,the spectrum,texture and shape features of the images are synthetically used to construct the object semantic description and establish merger rules in the process of region merging.Nonsubsampled contourlet transform and fuzzy clustering analysis are used to calculate the texture distances of the objects.The global optimization strategy is adopted to guarantee the multi scale characteristic of the method in the merging stage,and the region adjacency graph and nearest neighbor graph are used to maintain the adjacency relations which can ensure the efficiency of the algorithm.By comparing with other algorithms,it is proved that the algorithm can match the same kind of commercial software in accuracy and efficiency.(3)Asupervised evaluation method of optimal segmentation scale for remote sensing images is proposed.This method constructs the evaluation function of the scale according to the similaritiesbetween the reference objects and the segmentation objects,and the optimal segmentation scale is determined by calculating the value of the evaluation function.The inaccurate description of gray similarity is overcomed by the using of statistical histogram,and the geometric differences between objects can be made more accurate judgmentby constructing a shape description function to calculateshape similarity.The method is compared with the mature artificial method through experiment,and the validity of the method is testifed.(4)An object based road extraction method for high resolution remote sensing image is developed.The optimal scale segmentation result is firstly obtained by applying the segmentation algorithms and the optimal segmentation scale evaluation method proposed in this paper,and then the initial road extraction results are achieved by constructing the road knowledge base and computing the object features.At last multi direction morphological filtering is used to the initial extraction results to remove the similar mixed objects and optimize the road extraction results,and the optimized results are refined and connected so as to realize the extraction of road network.Experiment results show that the proposed method can preferably extract road network from complex remote sensing images.