Texture- and threshold-based image segmentation for quantifying Asian soybean rust severity in digital images
TL;DR
An automated image segmentation pipeline is developed and validated, implemented in Python, to quantify ASR severity from images of soybean leaflets, and enhances contrast between healthy and symptomatic tissues using color space transformations, histogram equalization, and gamma correction.
Abstract
Asian soybean rust (ASR), caused by Phakopsora pachyrhizi, is among the most destructive fungal diseases of soybean worldwide, making accurate severity quantification essential for breeding programs and disease management. In this study, we developed and validated an automated image segmentation pipeline, implemented in Python, to quantify ASR severity from images of soybean leaflets. The method operates in two stages: the first enhances contrast between healthy and symptomatic tissues using color space transformations, histogram equalization, and gamma correction; the second performs lesion segmentation based on Haralick texture features as a decision criterion and Otsu’s automatic thresholding. Validation was conducted on a subset of 50 images covering severity levels from 0 to 72.3%, using ImageJ pixel-level annotations as the reference standard. Agreement between the proposed method and reference measurements yielded a Lin’s concordance correlation coefficient (CCC) of 0.99, with deviations within ± 3%, confirming the accuracy of the approach. The pipeline processes hundreds of images in approximately 20–25 min.