Remotely sensed imagery intelligent interpretation based on image segmentation and support vector machines

Abstract

Remote sensing provides a useful source of data from which updated land cover information can be extraction for assessing and monitoring environment changes. This paper aims at achieving improved land cover classification performance based image segmentation and support vector machines (SVMs) classification. The object-based classification approach overcame the problem of salt-and-pepper effects found in classification results from traditional pixel-based approaches. The proposed method is a three-stage process, which makes use of the object information from neighboring pixels. Firstly, a robust image segmentation algorithm is used to achieve more homogeneous regions. Secondly, feature information is extracted from each segment and training samples is interactive selected in geographical information system platform. Thirdly, support vector machines classifier is employed to classify the land covers. The experimental results indicate that improved classification accuracy and smoother (more acceptable) is achieved compare with the traditional pixel-based method. Because of the image segmentation process significantly reduces the number of training samples, make SVMs classification method can be applied to information extraction from remotely sensed data.