The Research On Land-Cover Classification Based On Multi-Instance Learning In Haidian,Beijing

Land cover reflects the situation of land cover,and all human production and life activities are based on this.In recent years,land cover research has been a hot spot in the study of global environmental issues.With the rapid development of remote sensing technology in China,high-resolution satellites such as GF series have appeared,making it possible to achieve large-scale mapping and monitoring of land cover.Using remote sensing technology to obtain land use / land cover classification has become an effective tool with technical means.The development of remote sensing image acquisition technology urgently needs the level of remote sensing information processing and classification.With the rapid development of machine learning and deep learning in recent years,more and more machine learning algorithms are used for automatic classification of remote sensing images.Support vector machine is the most commonly used method of machine learning and pattern recognition since the new century.It has the characteristics of simple structure and good adaptability.It can solve problems such as high-dimensional space,nonlinearity and over-learning.It is used by many scholars.For remote sensing classification of land cover.Support vector machines have good applicability in the field of automatic classification of remote sensing,but there are still problems such as insufficient precision and a large amount of time cost caused by complex operations.This paper takes Haidian District of Beijing as the research area,based on the traditional support vector machine algorithm and the idea of sequence minimum optimization(SMO)and multi-instance learning,proposes an improved algorithm(MISMO),and implements automatic land cover for the research area classification.This article has mainly achieved the following results:(1)Based on the remote sensing image data of GF-2,the actual classification data is obtained by means of human-computer interactive interpretation,and the land cover classification system in the study area is established by combining the status of the study area and the field verification data.It is divided into seven categories: cultivated land,garden,forest,river,other waters,construction and unused land.(2)This article combines the sequence minimum optimization algorithm to improve multi-instance learning.Use the improved algorithm MISMO to solve the problem of automatic land cover classification.After automatic classification of the entire study area,the overall accuracy of the classification in the study area was 93.65%,and the Kappa coefficient was 88.39%.The classification results and accurate evaluation results are reliable.(3)This article selects some areas in the study area and compares the improved algorithm with the traditional support vector machine algorithm(SVM)and package-level multi-instance support vector machine algorithm(MISVM)automatic classification results.Both are superior to traditional algorithms.