An Integrated System for Semi-automated Segmentation of Remotely Sensed Imagery

The recent trend in image classification of agricultural areas is towards the use of per-field approaches that work in a manner that a crop label is assigned to each field. The existing field boundaries are commonly used as ancillary information for performing fieldbased image classification. However, the presence of multiple crops within the fields may cause problems during per-field image classification. To obtain more reliable results, within field image segmentation is needed to be performed for detecting sub-field boundaries. In this study, an integrated system was developed for performing semi-automated segmentation of remotely sensed imagery. The system works based on a field-based logic to perform image segmentation within agricultural fields in an integrated environment of Geographic Information System (GIS) and Remote Sensing (RS). The agricultural fields are selected through a database query and within field segments are detected. First, an edge detection process is performed on the image and the detected edges are vectorized to extract the straight line segments. Then, these line segments are correlated with the existing field boundaries through perceptual grouping and the closed sub-fields are formed. The sub-fields represent within field segments, each of which contains a distinct crop type. The proposed approach was implemented using a 10-m resolution SPOT5 HRV image and a 20-m resolution SPOT4 HRV image covering a part of Karacabey Plain, Turkey. The results illustrate a high degree of agreement with the reference data.