Development of a Nonparametric Active Contour Model for Automatic Extraction of Farmland Boundaries from High-Resolution Satellite Imagery

Abstract Agricultural field maps are significant sources of data to achieve precision farming. The present research is a step toward generating land use/land cover maps automatically. The primary goal of this research was to develop an area-based model from nonparametric active contour models for agriculture land boundary extraction from IRS P5 satellite images. After investigating two well-known models created from nonparametric active contours, named local binary fitting and multi-phase, the local binary fitting model was selected to develop and enhance. Land boundary detection was improved by adding two texture layers to the input images and the development of the external energy function. The local binary fitting model was advanced as a multi-phase model in order to identify several regions in an image. Also, dull image boundaries were better extracted by changing the sigma parameter and regularization term. Evaluation of the proposed method yielded to the overall accuracy, user accuracy, producer accuracy, and kappa coefficient of 89.53%, 65.93%, 86.13%, and 86.52%, respectively.