Robust Vegetation Segmentation for Image-based Field Survey

Abstract

This paper proposes a robust vegetation segmentation method against illumination changes and noise, using a unique likelihood with efficient optimization and explainable thresholding for stable image-based field surveys. The proposed segmentation provides crop-focused measurements, with small variance, with plant cover maps. Analysis concerning the observation angle indicated the effectiveness of a stable and efficient survey when wide-angle cameras are used. This benefit is provided by filtering out visible soil pixels, whose proportion per unit area varies with the observation angle. Comparison with the conventional soil-adjusted vegetation index (SAVI) illustrated the achievements for robust segmentation under fickle outdoor illumination.