Research Of Land Cover Classification In China Based On FY-3A MERSI Data

Land cover data is the basis data of earth science research, it has important applications in geography, atmospheric science, agronomy, forestry and other fields. Land cover classification is a fundamental work in land cover research. Using of remote sensing technology to obtain detailed, accurate, real-time land cover data has become hotspot in large-scale land cover research. China has independently developed a new generation of polar-orbiting meteorological satellite FY-3A. It loads Medium Resolution Spectral Imager (MERSI) having a plurality channels of the visible, near infrared, thermal infrared, providing an important data source to identify of large-scale land cover.In this paper, we use MERSI reflectance, vegetation index, brightness temperature and DEM data, using Maximum Likelihood, BP Neural Network and Support Vector Machine as the classification method. In the five experiment area of Jilin Province, Jiangsu Province, Hubei Province, Qinghai province and Guangdong Province, to carry out pre-test for land cover classification.The main conclusions are the following:(1) According to the characteristics of MERSI data, the first five bands with250m resolution, The one to five band reflectance, vegetation index, brightness temperature data and DEM data as the classification feature, it can effectively select samples, then can faster and better classify land cover in the study area.(2) Classifier selection is important to improve the recognition accuracy. In this paper, using the Maximum Likelihood, BP Neural Network and Support Vector Machine to classify land cover based on MERSIL1data in five test area of Jilin Province, Jiangsu Province, Hubei Province, Qinghai province and Guangdong Province. From the matrix confusion can be seen, Support Vector Machine has the highest recognition accuracy. So, Support Vector Machine is used as classifier to classify land cover in the China area.(3) Based on MERSI NVI vegetation index products, choice data of February, April, August, November in2012, Support Vector Machine was used to study land cover classification, finally, got four different seasons results, and using comparison analysis and crop ground point to verify the accuracy. From the area consistency analysis, the classification results for four months with VIRR land cover product overall consistency, the correlation coefficient of four months are very high, the highest is April, R2is0.9294; From the confusion matrix analysis, the overall accuracy of four months maintain about eight, the highest overall accuracy of February is82.6268%, the Kappa coefficient is0.7962; From agricultural land type with crop ground sampling points of Chinese crop growth and soil humidity ten days dataset, agricultural land type keep good agreement with crop ground sampling points, maintained at around90%. Indication of results of accuracy verification, MERSI data and Support Vector Machine can be used in large-scale land cover classification.