Research On Object Oriented High Resolution Remote Sensing Image Classification Algorithm

With the development of remote sensing technology in recent years,the resolution of remote sensing image is also improved.High-resolution remote sensing image has been widely used in various industries and has become a hot topic in remote sensing image research.Among them,the classification of remote sensing images can effectively extract the classification information of ground objects,which is the key link of remote sensing image processing and interpretation,and has attracted the attention of a large number of scholars.However,the improvement of resolution not only enriches the ground object information in the image,but also increases the amount of data in image processing,making it difficult to fully apply the pixel-based classification method.Object-oriented classification method based on the homogeneous pixel set for implementation,to a certain extent,solve the problem of large amount of data,at the same time it can effectively extract the texture in the pixels,and the characteristics of space,make full use of the rich feature information,gradually become the mainstream of the remote sensing image classification method,its on the land use and land cover test with high research value.Therefore,this paper focuses on the key problems in object-oriented classification methods,and proposes two new object-oriented classification algorithms.The main research work is as follows:(1)Aiming at the problem that the traditional simple linear iterative clustering(SLIC)super-pixel segmentation algorithm is difficult to accurately segment small objects,and study and put forward a kind of super pixel image segmentation algorithm based on Canny edge operator,the algorithm using Canny edge operator to extract the feature edge information as a guide,to assist the SLIC algorithm for segmenting super pixels,and makes the boundary of the pixel block of generating the edge of the joint at terrain target.Experiments were conducted on the image data of GF-1 and GF-2,and the experimental results were analyzed and evaluated to verify the feasibility and effectiveness of the proposed algorithm.(2)An improved object-oriented classification algorithm is proposed.In the traditional object-oriented classification algorithm,the classification effect of the same ground object category under different classification algorithms is inconsistent;In the same classification model,the classification accuracy of different types of ground objects varies greatly.In order to solve this problem,firstly,support vector machine(SVM)and random forest(RF)classification models are used for classification.For one or more ground object information,the classification algorithm with the highest classification accuracy is used to extract such ground object information.By integrating all ground object information,remote sensing image classification is finally realized and the classification accuracy is improved.Experiments were conducted on the image data of GF-1 and GF-2,and the experimental results were analyzed and evaluated to verify the feasibility and effectiveness of the proposed algorithm.