Colour Segmentation in Coffee Berry recognition: Light and Texture Convergence for Single-Overflight Frugal Agro-Diagnostics in smallholder coffee environments

Abstract Smallholder coffee plantations are disadvantaged compared to larger coffee estates because they lack the digital sophistication to perform regular investigation of crops for signs of berry disease and poor berry health. Accurate visual recognition of coffee berry health is critical for early detection, particularly by pests such as Hypothenemus hampei . A persistent challenge is distinguishing true signs of infestation, and the pursuit of healthy berries is a contested struggle, classified by the size and scale, and determined by constrained financial decision support systems preventing high-level precision crop evaluation. Much of the precision agriculture discourse ignores smaller concerns in the 2-hectare (ha) to 5-hectare (ha) sectors. In Vietnam, Indonesia, Colombia, Uganda, and Ethiopia it is vital to pursue optimisation strategies for smallholder coffee growers. This addresses the need for practical improvement practices in the 2-5 ha range. This paper presents a non-neural network image process that integrates colour segmentation, specular highlight analysis, and texture convergence using single overflight diagnostics. The study posits a frugal, action-oriented methodology combining a single low-altitude drone overflight with a smartphone-based participatory ground truth to generate predictive, map-based risk triage for smallholder growers. The approach uses Action Design Research for grower-centric remote sensing. The method treats bacterial/viral detection as symptom-level triage, enforces confirmatory sampling, and iteratively adapts via active learning, all within a low-cost app architecture suitable for smallholder ≤5 ha plantations and ≤2 kg drones. The multi-stage pipeline integrates this into a rule-based decision system, capable of classing berry health without reliance on deep learning. This approach offers a robust, interpretable framework for agricultural applications, particularly in resource-constrained environments where neural networks are impractical. By leveraging light and texture convergence, the method provides for real-time, non-invasive coffee berry health monitoring.