Change Detection Analysis of Forest Areas Using Satellite Stereo Data

Tree height is a fundamental parameter for describing the forest situation which can be determined by different means. In this paper a novel region based forest change detection method is proposed using panchromatic CARTOSAT-1 stereo imagery. In the first step, DSMs from two dates are generated based on a new dense matching methodology called semi-global matching. To achieve reliable change detection a multi-step procedure has been developed using a combination of image data and DSMs. After 3D co-registration of the two DSMs, the othorectificated images are generated based on these DEMs. Mean-shift segmentation is applied to the ortho-images to get the initial regions. Following, the height change is extracted as well as grey value changes based on a region based level. The fusion of the several kinds of change sources are performed under the Dempster-Shafer statistical theory. To further improve the change detection result, texture measures called Grey Level Co-occurrence Matrix (GLCM) features are derived with changing window size and displacements and are analysed to extract the real forest change area. The test data are acquired over a forest area close to Freising, Germany, which consist of two pairs of stereo data from the year 2008 and 2009. Evaluation of the proposed approach proves its efficiency and accuracy.