RGB-Derived Haralick Edge-Texture Features Discriminate Mung Bean Responses to Mineral and Biofertilizer Regimes

Low-cost RGB imaging can accelerate crop phenotyping, but its ability to capture plant responses to fertilizer regimes through edge-based texture descriptors remains insufficiently tested. Mung bean [Vigna radiata (L.) R. Wilczek] was grown in a randomized complete block design with six fertilizer regimes and four replicates. Side-view RGB images acquired under controlled conditions were segmented, converted to grayscale, transformed into Canny edge maps, and summarized using six gray-level co-occurrence matrix (GLCM) features. Agronomic variables were evaluated: plant height, stem diameter, nodule weight, fresh weight, pod weight, shoot dry matter, grain yield per plot, the chlorophyll index, root length, and root volume. Contrast, energy, homogeneity, entropy, and variance differed across regimes, whereas the GLCM correlation did not. Mineral fertilization resulted in the highest values of stem diameter, nodule mass, fresh biomass, shoot dry matter, and the chlorophyll index. Baker’s Gamma coefficient (0.9369) and the cophenetic correlation (0.9498) showed high similarity between the hierarchical structures, indicating that the texture descriptors preserved a treatment organization highly similar to that observed for the agronomic variables. The Haralick features calculated from controlled RGB-derived edge maps revealed fertilizer-associated differences in the mung bean canopy structure. This method is promising for nondestructive screening, but replicate-level predictive validation, full reporting of image-processing parameters, and external testing are needed before agronomic traits can be estimated from images.