A Comparison of Artificial Neural Networks and Support Vector Machines on Land Cover Classification

Artificial Neural Networks (ANNs) as well as Support Vector Machines (SVMs) are very powerful tools which can be utilized for remote sensing classification. This paper exemplifies the applicability of ANNs and SVMs in land cover classification. A brief introduction to ANNs and SVMs were given. The ANN and SVM methods for land cover classification using satellite remote sensing data sets were developed. Both methods were tested and their results of land cover classification from a Landsat Enhanced Thematic Mapper Plus image of Wuhan city in China were presented and compared. The overall accuracy values of ANN classifiers and SVM classifiers were over than 97%. SVM classifiers had slightly higher accuracy than ANN classifiers. With demonstrated capability to produce reliable cover results, the ANN and SVM methods should be especially useful for land cover classification.