Landforms' Classification Of Islands And Reefs In South China Sea Based On SVM

China is a large maritime country with many islands and reefs,which have very important values.But the situation of China's islands and reefs is worrying.For example,many islands and reefs in the South China Sea have been illegally occupied by other countries in various ways,and this has sounded an alarm to us.In order to protect the legitimate rights of China's oceans and investigate the status of islands and reefs in the South China Sea,China Aero Geophysical Survey and Remote Sensing Center undertook the project of "Remote Sensing Survey of Basic Geology of Islands and Reefs in the South China Sea" in 2014.A sub-task of the project needs to classify various landforms of islands and reefs.The main method currently used is visual interpretation using remote sensing images.But this method is inefficient.In order to solve this problem,this paper uses support vector machine and other methods to classify landforms of Boji Jiao and Royal Captain Reef.The final result is good.In this paper,we mainly use ISODATA algorithm,maximum likelihood algorithm and support vector machine to classify landforms of islands and reefs in the South China Sea.This paper focuses on the research of support vector machine.This method uses the kernel mapping,which overcomes the difficulty of dimensional over-growth and solves the non-linear problem well.And because the kernel function and the parameters have a great influence on the result of the support vector machine algorithm,this paper uses the grid search to optimize the parameters of the four kernel functions(linear,polynomial,sigmoid and gaussian kernel)respectively.Finally,the optimal classification kernel function and parameters are determined.Then the landforms of islands and reefs are classified.Finally,I use the results of visual interpretation as the basic data,calculate the confusion matrix of the above classification results,and the corresponding classification accuracy is obtained.The accuracy of ISODATA and maximum likelihood algorithm is about 73%.The accuracy of linear kernels,polynomial kernels and Sigmoid is about 85%.Classification accuracy of support vector machine based on RBF Gaussian kernel is about 91%.This accuracy is very high,and generally better than other methods and kernels.This also shows that this method is suitable for the project of "Remote Sensing Survey of Basic Geology of Islands and Reefs in the South China Sea",and SVM can be used to classify the landforms of islands and reefs in South China Sea.