Crop field extraction method based on texture analysis and automatic threshold determination

We developed a crop field extraction method, which is one of the steps in our land cover classification system for a huge area, like national or continental scale. Firstly, a crop field model is defined to characterize “crop field” in terms of NDVI value and textural information. Textural information is represented by the density of straight lines which are extracted by wavelet transform. Secondly, candidates of NDVI threshold value are determined by the “scale-space filtering” method. The most appropriate threshold value among the candidates is determined by evaluating the line density of the area extracted by the threshold value. Finally, the crop field is extracted by applying level slicing to Landsat TM images with the threshold value determined above. The experiment demonstrates that the extracted area by this method coincides very well with the one extracted by visual interpretation