Sub-pixel land cover classification using support vector machines

Over the last few years, support vector machines (SVMs) have shown a great potential as classifiers for remotely sensed data. Generally, these have been used to perform conventional hard classification where each pixel is allocated to only one class. Remote sensing images, particularly at coarse spatial resolutions, are contaminated with mixed pixels that contain more than one class on the ground. Hard classification process may result in erroneous classification of images dominated by mixed pixels. Therefore, sub-pixel classification approaches that decompose the pixel into its class constituents in the form of class proportions have been advocated. In this paper, we propose a SVM based algorithm for sub-pixel land cover classification. The proposed SVM based algorithm uses probability estimates for multiclass classification by pairwise coupling. The algorithm is employed to produce sub-pixel land cover classification from a Landsat ETM+ image. Classification accuracy achieved is assessed using three measures, namely, the overall accuracy obtained from a fuzzy error matrix, the squared correlation coefficient, and the root mean squared error. The results are compared with the posterior probabilities derived from the maximum likelihood classifier (MLC) and the fuzzy classification based on MLC. Our experiments show that accuracy obtained from the proposed algorithm is significantly higher than the two bench-marked classifiers. Thus, the outputs from SVM based algorithm can be used to reflect the actual class composition of the pixels on ground.