Research On Semantic Segmentation Method Of High Spatial Resolution Remote Sensing Image

With the continuous development of aerospace platform and remote sensing technology,we can obtain more and more high spatial resolution remote sensing image with clear texture details and rich spectral features,High spatial resolution remote sensing image has become an important spatial information source in economic development,national security,geographic information service and other aspects.At the same time,the phenomenon of " same object with different spectrum" and " different objects with same spectrum " followed,and became the main factors restricting the application and development of high spatial resolution remote sensing images.In order to solve the above problems,many scholars have done a lot of research at home and abroad.At present,the analysis and processing of high spatial resolution remote sensing image has become one of the hottest directions in the field of remote sensing.Meanwhile,with the improvement of image spatial resolution,information extraction technology is moving from pixel classification to the object oriented recognition,Object oriented image analysis method uses the mode of first segmentation and classification to identify and extract objects.Among them,the technology of remote sensing image segmentation and classification are the foundation of the object-oriented image analysis method.The two processes involve the selection of features and the calculation of feature similarity,which will bring the uncertainty of the results in the two processes.With the deepening of computer vision research,researchers pay more attention to the accurately analysis and understanding of image segmentation,the proposed semantic segmentation for this problem has brought opportunities,which is different from the traditional image segmentation method,the fundamental purpose of semantic segmentation is set in the semantic category of segmentation,this is of profound significance to the more accurate image analysis and processing.Therefore,we take the high spatial resolution image as the research object in this paper,and use the image segmentation method and object based image analysis method to study the new method of semantic segmentation.(1)High spatial resolution remote sensing image have more texture details and the enormous amount of data,the study poses a great challenge to the remote sensing image semantic segmentation,comparing the advantages and disadvantages of various segmentation methods,analysis of the research status,key technologies and problems.(2)In this paper,a semantic segmentation method combining wavelet domain multiscale and Markov random field(GMGRF)are proposed.Among them,the marker field is modeled by MLL model,the potential function is estimated,and then the Gauss model is used to model the wavelet coefficients vector field on each scale,and the expectation maximum algorithm is used to obtain the GMGRF model parameters.The overall precision of the semantic segmentation of the high resolution remote sensing image of the Apeng village is 78.7073%,and the overall accuracy of the semantic segmentation of the high resolution remote sensing image of Dema village is 89.7524%.(3)Research on semantic segmentation method based on simple linear iterative clustering(SLIC),in view of the method,it improves the segmentation efficiency,at the same time,the over segmentation problem is serious,In this paper,according to the region merging algorithm,the phenomenon of over segmentation is solved by combining the spectrum,texture,shape and area,and semantic segmentation are realized by labeling the semantic category according to the mean of the spectral.There are three groups of experiments in the high resolution remote sensing image of Apeng village,The whole precision of the first experiment is 78.7802%,the whole precision of the second experiment is 80.1948%,and the whole precision of the third experiment is 77.8281%.There are three groups of experiments in the high resolution remote sensing image of Dema village.The whole precision of the first experiment is 88.2376%,the whole precision of the second experiment is 88.4100%,and the whole precision of the third experiment is 88.7041%.(4)The image segmentation algorithm based on Markov random field model is studied.Modeling of the image on the basis of MRF,according to the condition of iterative model and gradient minimization model,proposing the improved iterative segmentation algorithm based on the model of semantic conditions.Through comparative experiments,the overall accuracy of the semantic segmentation of the high resolution remote sensing image of the Apeng village was increased by 2.7086%.The overall accuracy of the semantic segmentation of the remote sensing image of the Dema village was increased by 2.6333%.(5)The accuracy evaluation and analysis of different experiments are carried out.