Study on Methods Comparison of Typical Remote Sensing Classification Based on Multi-temporal Images: Study on Methods Comparison of Typical Remote Sensing Classification Based on Multi-temporal Images

The study area is located in Miller county,Yunnan province,China.An experiment to select the appropriate classification method for multi-temporal remote sensing images was done.Typical classification methods including Object-Oriented Classification(OOC),Back Propagation Neural Network(BPNN),Spectral Angle Mapper(SAM),Maximum Likelihood Classifier(MLC),and Comparison After Classification(CAC)were tested in this experiment.In this study,using two-phase remote sensing images of CBERS02B-CCD and Landsat-5 TM,the suitability and accuracy of typical methods to deal with multi-temporal images classification were compared,based on different phenological characteristics of sugarcane,corn and paddy.Using full sample test method,visual interpretation results were used as reference data to validate the accuracy of different classification methods.The experimental results show that the order of overall classification accuracy from high to low is OOC,BPNN,SAM,MLC,and CAC,and the Kappa accuracy of them is 0.655、0.635、0.631、0.601 and 0.577,respectively.As it is easy to identify paddy,its accuracy is higher than that of sugarcane and corn.The order of accuracy of paddy for different methods is as the same as the order of overall accuracy,the highest and lowest accuracy of paddy is 0.706 and 0.621,respectively.The accuracy curve position between the accuracy of various land covers and the overall accuracy are consistent for MLC and CAC,and the overall accuracy of CAC is the lowest one.The accuracy of corn for OOC is the highest one with Kappa of 0.611.The Kappa accuracy of sugarcane for OOC,SAM and BPNN is 0.594,0.575 and 0.575,respectively.In general speaking,for the remote sensing classification of Multi-temporal Images,OOC is the best,BPNN and SAM is better,MLC and CAC are the worst.The conclusions of this experiment have some guidance to select the appropriate classification method for multi-temporal remote sensing images.