Detecting regular plantation areas in satellite images

Detecting, segmenting and classifying agricultural fields in remote sensing images enable advanced planning of the land use economically. As most human structures, plants are cultivated in some order in orchards or farms. In this paper, a regularity detection method is proposed for exploiting this order information. The method slides windows over the spot filter responses of satellite images and analyzes their projection vectors. A regularity coefficient is calculated for each window. These regularity coefficients are further used for creating a regularity map, where regular regions obtain higher scores. These regularity maps can later be employed for the segmentation and classification of cultivation lands. The proposed method is illustrated in the detection of hazelnut orchards in sample high resolution satellite images.