Study Of Remote Sensing Image Classification Based On Support Vector Machines

With the rapid development of computer technology, automatic classifications ofremote sensing images with computer began to be taken seriously, and these methodshave greatly improved the quality, accuracy and efficiency of image classification.However, classification and interpretation of remote sensing image based on thetechnology of traditional remote sensing image classification method is the use ofstatistical laws of remote sensing image`s pixel values, whose classification accuracyof the results is limited by the number of training samples and classification modelalgorithm.Support vector machine, established on the basis of the VC dimension ofstatistical learning theory and structural risk minimization principle, is one of machineclassification methods with a small amount of sample cases. It has shown greatapplicability in the actual applications of remote sensing image classification. In thispaper, under the auspices of the Sino-German scientific and technological cooperationand exchange programs (2007DFB70200) and Shandong Province Natural ScienceFoundation (Y2008E10), remote sensing image classification method based onsupport vector machine has been discussed.In this paper, I have used a method of support vector machine classificationcombined with image spectral characteristics extraction for remote sensing imageclassification research. In order to choose an optimal classification model, fivecategories support vector machine classification model and three sets of samples withdifferent number have been selected for classification experiment. Finally, comparethe classification results with the traditional classification of machine classificationalgorithm results. The comparative study show that result of support vector machine classificationunder the conditions of the three groups of training samples showed highclassification accuracy, whose overall classification accuracy higher than 85% andKAPPA coefficients were greater than 0.8; Under the case of changes in the numberof samples, The classification accuracy of support vector machines for classificationchange is the smallest, showing good adaptability.During the study work, I have found that support vector machine classificationalgorithm can be well adapted the training and classification of remote sensing imageswith a small amount of sample cases; this method will show great adaptability todifferent classification tests by adjusting the kernel function according to differentsituations; it has a certain advantage of algorithm and can effectively avoid thephenomenon of "over learning". As an important method in the remote sensing imageclassification, a depth understanding of the advantages and disadvantages in theapplication of image classification support vector machine has an important guidingsignificance in the future study of remote sensing technology.