The technology of agricultural fields remote sensing image segmentation using morphological profiles

Abstract

Automatic agricultural monitoring requires field borders to be known a priori. Although geoinformation systems contain vector field maps, the annual sowing pattern changes lead to noticeable field border changes. That is why remote sensing image segmentation is widely applied to achieve current field border estimation. However, agricultural remote sensing image segmentation becomes a very challenging task because of an oversegmentation. Oversegmentation consists in a multiple segments formation within one semantically homogenous region and usually arisesfrom local brightness variations caused by relief irregularities, soil tillage variations and local differences in crop growth. The aim of our study was to developa segmentation technology that would be able to reduce oversegmentation. The technology proposed in this paper includes three stages: feature extraction, segmentation and post-processing. We suggest using any multichannel image segmentation algorithm with two additional steps that might lead to oversegmenation decrease. The first step is feature extraction, which consists in morphological profiles of the normalized difference vegetation index (NDVI) calculation instead of applying simple brightness features. The second step is post-processing procedure, which controls the anthropogenic shape of parcels within the field.To evaluate the decrease in oversegmentation produced by our technology, we chose the basic segmentation algorithm and compared the quality of basic segmentation algorithm (using simple brightness features and without post-processing) with the results of the proposed technology (using morphological profiles and post-processing). The experimental results have shown that proposed technology provides more visually suitable segmentation and reduces the oversegmentation by 3% for morphological profiles with five components. The results of our investigation might be applied in different applications for automatic agricultural monitoring using remote sensing data.