Land cover change detection based on multi-temporal Spot5 imagery

In this paper, we propose a land cover change detection method especially for the change between farmland and non-farmland usage, which is of vital importance in urban monitoring application. Our approach is based on the difference of NDVI (Normalized Difference Vegetation Index) and texture characteristic between farmland and non-farmland in SPOT-5 imagery, Spot5 imagery is used because it has high spatial resolution and shorter recursive period. First, NDVI is used for a rough detection of land cover change. Since NDVI is an important parameter to describe land cover, NDVI difference between two different temporal images can be calculated through image operation to obtain change information. After that, unsupervised texture segmentation of the image of NDVI difference is performed to find a detailed land cover change. During the process, Gabor filter is used for the description of texture features and k-means clustering algorithm is proposed to cluster those pixels. Finally, areas which have different textures are land cover change regions. A suburb area in Beijing is selected to verify our approach; two spot5 images in different time are used. And the experiment result shows that the change detection accuracy related to non-farmland usage is more than 80% comparing to result interpreted by expert. The creative parts of our approach lie in a three-step detection scheme, and NDVI is proved to be an efficient and effective way to differentiate the change area. Although our approach is successful in our case, it is not suited for remote sensing images captured in winter as the NDVI and texture difference in this case is minors.