A New Smoothing-Based Farmland Extraction Approach with Vectorization from Raster Remote Sensing Images.

With the increasing resolution and application scene of remote sensing images, land and resource investigators begin to consider using these images to investigate crop species, cultivated area and ownership. Instead of manually drawing the boundary of the selected farmland region, an efficient edge-preserving smoothing method for automatically segmenting and extracting the area is proposed, which is performed according to the following three steps: (1) Remove the interference information by image preprocessing. The smoothing algorithm in this process was proposed according to features of the ideal smoothed image and using Maximum a Posteriori estimation model to preserve borderlines of farmland regions; (2) Image segmentation, including threshold and region segmentation using the fixed threshold and the hole removal based on the region growth method respectively after edge and whole image enhancement; (3) Information extraction, including region separation with the Flood Fill method and region vectorization which can reduce the amount of data and make the image to scale arbitrarily by contour tracking after thinning with the Freeman Chain Code. The final results of segmenting and extracting farmland objects with different features from raster remote sensing images demonstrate the correctness and efficiency of the proposed process.