Severity-based analysis of cassava leaf diseases for precision agriculture using image processing

Abstract

Cassava is a critical food security crop in many tropical regions; however, its productivity is constrained by foliar diseases, including bacterial blight, brown spot, green mite infestations, and cassava mosaic disease. While accurate and timely assessment of disease severity is essential for effective crop management, most existing image-based approaches focus on disease detection rather than quantitative severity estimation. This study presents a lightweight image-processing framework for pixel-level estimation and comparative analysis of cassava leaf disease severity using RGB (Red, Green, and Blue) images. The framework segments cassava leaves using HSV (Hue, Saturation, Value) colour thresholding, identifies diseased regions through clustering in the LAB (Lightness, A-channel, B-channel) colour space, and quantifies disease severity as the ratio of infected pixels to total leaf area. Severity levels are classified as low, medium, or high to support agronomic decision-making. Experimental results across four cassava leaf conditions demonstrate distinct and quantifiable differences in disease severity. Bacterial blight had the highest mean severity (11.3%), indicating extensive infection, followed by brown spot (9.8%), while green mite and mosaic showed lower mean severities (8.5%). Analysis of severity levels showed that infections in the dataset were limited to low and medium categories. Bacterial blight accounted for most medium-severity cases, while green mite and mosaic were mainly mild. The proposed framework is computationally efficient, interpretable, and suitable for resource-limited farming environments. By providing quantitative severity metrics and spatial heatmaps rather than basic disease labels, it supports evidence-based prioritisation and precision crop management.