A Comparative Study Of Remote Sensing Image Classification Method Under Different Geomorphic Unit

The development of remote sensing technology enables us to obtain abundant information.The choice of classification method is a key link, and how to identify the multidisciplinary images and reach to a certain degree of accuracy is a key problem in the study of remote sensing images. Choosing reasonable classification methods will not only improve the classification accuracy,which is of very vital significance.Taking hongtong county in Shanxi Province as the study area, and Landsat8 OLI remote sensing image on March 21, 2014 as data sources, the study uses maximum likelihood method(MLC), artificial neural network(ANN) and support vector machine(SVM). The remote sensing images under different geomorphic units in the studied area are classified, and the classification accuracy is analyzed comparably.The classification method suitable for different geomorphic units is determined by the accuracy evaluation, which provides reference for quick access to land use/land cover data of study area in future. The main research conclusions are as follows:(1) Using maximum likelihood method(MLC), artificial neural network(ANN) and support vector machine(SVM), the study extracts information for land use/land cover classification in study area. It shows that using support vector machine(SVM) and artificial neural network(ANN) to classify remote sensing image can effectively improve the classification accuracy, and classification accuracy is much higher than that of maximum likelihood method(MLC). The classification accuracy of SVM method is higher than that of ANN and that of MLC. The classification results of SVM is close to the actual value, and can distinguish land use type better, improve the accuracy of the information of land use, and can be used as an effective method for remote sensing image classification.(2) Under different geomorphic units, three kinds of classification is applied in plain and hilly area. The classification results show that the support vector machine(SVM) classification method is superior to the maximum likelihood method(MLC) and artificial neural network(ANN) classification. The classification accuracy of SVM for forest land in hilly area is much higher than that in plain; as for arable land and construction land, the classification accuracy for plain is higher than that for hilly area. Thus, the SVM classification method can be the optimization for information extraction and monitoring of different geomorphic units.(3) Under different sample unit, different classification methods lead to different remote sensing image classification results. It is found that as for remote sensing image classification under large sample unit, support vector machine(SVM) and artificial neural network(ANN) need more time to learn; as for remote sensing image classification under small sample unit, they need less learning time,and support vector machine(SVM) method needs less learning timethan artificial neural network(ANN) and has higher classification accuracy. The overall accuracy of three methods of remote sensing image classification: Classification accuracy under large sample units is lower than that under smaller sample unit.(4) Under the small sample unit, the improved artificial neural network can effectively increase the accuracy of the land type classification, and the accuracy is higher than that of MLC. The classification accuracy of the improved artificial neural network has reached to more than 93.22%, and the Kapp coefficient reached to0.8674. The classification result is most close to the actual value,which can distinguish land use type better.(5) Through the above analysis, support vector machine(SVM)classification is the optimization for the remote sensing image classification under different geomorphic units. Remote sensing image classification in hilly area: Artificial neural network(ANN) is the optimization for the cultivated land, support vector machine method(SVM) is the optimization for forest land and construction land,and maximum likelihood method(MLC) is the optimization for other construction land. Remote sensing image classification in plains:Support vector machine method(SVM) is the optimization for the cultivated land and waters, maximum likelihood method(MLC) is the optimization for woodland, and artificial neural network(ANN) is the optimization for construction land. Support vector machine(SVM) is used for the remote sensing images under small sample unit, its classification accuracy is high, and the classification time is short.Therefore, according to different geomorphic units and different classification method, different optimization for information extraction and monitoring of different land type is chosen.