Developing an Integrated System for Extracting the Sub-fields within Agricultural Parcels from Remote Sensing Images

An integrated system was developed to detect sub-fields within agricultural parcels from remote sensing images. The permanent field boundaries are known and stored in a database. The detection of the sub-fields is carried out on field-by-field basis. The fields are selected one-by-one through a database query. The remotely sensed image is filtered using the Canny edge detector. The detected edges are then vectorized to generate the straight line segments. To reduce the number of line segments, a line simplification algorithm is carried out. Next, the line segments are associated with the existing field boundaries using the perceptual grouping rules and the sub-fields are generated. To do that, for each pair of the line segments, the analysis operations are carried out and an iterative decision mechanism is used to modify the line segments when extracting the sub-fields. The approach was implemented in an agricultural area located near Karacabey, Turkey, where majority of the fields are rectangular shaped that affects field-based image segmentation. The data used include SPOT4 and SPOT5 images and existing vector field boundaries. For each image, (i) the first component of the Principle Component Analysis (PCA) bands, and the intensity image of the first three bands ([Band1+Band2+Band3] / 3) were used. Of the four segmentations, the SPOT5-PCA image provided the highest accuracy of 83.8%, while the accuracy of the SPOT5-Intensity image (82.6%) was also not significantly different. The accuracies of the SPOT4-PCA and the SPOT4-Intensity images were computed to be 78.8% and 76.2%, respectively. It was observed that the under-segmented fields were slightly more for the SPOT4 images than those obtained for the SPOT5 images.