Satellite image classification using sparse codes of multiple features

This paper presents a new method for satellite image classification. Specifically, we make two main contributions: (1) we introduce the sparse coding method for high-resolution satellite image classification; (2) we effectively combine a set of diverse and complementary features-SIFT, Color Histogram and Gabor to further improve the performance. A two-stage linear SVM classifier is designed for this purpose, which firstly generate probability vectors for each image with SIFT, Color Histogram and Gabor features respectively and then the generated probability vectors with different features are concatenated as the input features of the second stage of classification. In the experiment of satellite image categorization, we And that, in terms of classification accuracy, the suggested classification method using sparse codes of multiple features achieves very promising performances and the linear kernel can remarkably reduce the complexity of the SVM classifier.