Residential Area Extraction From Remote Sensing Imagery Based On Multi-classifier Combination

Residents are the human settlement place that formed in accordance with the needs of production and life. It is not only one of the important terrain elements, butalso an importantexpressed content of the topographic map.At present, when mapping topographic map with remote sensing image, the acquisition of attribute information of topographic elements is still relied on artificial visual interpretation, which exists the problem of labor intensity and low efficiency, and can be easily influenced by interpretation staff’s theoretical knowledge and practical experience. The method of using computer automatic interpretation to obtain the attribute information of residents is a difficult and hot issue in the study of remote sensing image interpretation, which has important theoretical significance and practical value for topographic map surveying and updating, and battlefield geospatial information system.Based on the study of neural network method,support vector machine(SVM) method and other classical classification algorithm, this paper uses multiple classifier combination method to adopt to the inhabitants of many kinds of remote sensing image to extract, focuses on the multiple classifier combination residents extraction strategy and multiple classifier combination optimization algorithm to extract, anduses the precision evaluation algorithm to quantitatively evaluate the classification extraction accuracy of residents extraction result.The paper tightly rounds the residents extraction of remote sensing image to launch the research. The main contents and innovations are as follows:1. Research status of residents extraction technology is summarized.On the basis of the analysis of characteristics of residents in remote sensing, it does experimentsfor a variety of classification algorithm, and bases on the residents of the classical classification algorithms to extract. Combined with the experiment, it analyzes and discusses the various classic classifiers, which can be the foundation of extracting multiple classifier combination residents, and residents to extract quantitative evaluation of the results are given.2. Existing multiple classifier combination methods which are applied to residents classification are studied, which are verified and analyzed by the residents classification experiments. Based on the detailed analysis of performances and characteristics of multiple classifiers, usingvoting method to select classifierfor composite applicationsand discuss the combination principles of base classifiers is applied.Multiple experiments are designed to determine the base classifier and classification rules.Relevant results are obtained.3. Category with maximum probability method and the fuzzy integral fusion method for multiple classifiers combination residents extraction algorithm is optimized. When the base classifier acquired different pixeltypes, the maximum probability category method is used to determine the category of pixels.And the fuzzy integral fusion method is a kind of weighted average of the promotion. After carrying on the ballot and maximum probability method execution, the algorithm can further improve the classification accuracy.4. The classification result obtained by classical classification algorithms and multi-classifier combination algorithm is raster image. Thereare small spots and voids,which cannotmeet the requirements of residential elements represented on the topographic map. It needsa series of post-processing to obtain the planar residents or residents planar contours. In this paper, mathematical morphology operation is used to remove spots for area residents, and the boundary tracking method is used to acquire the planar contour lines.