Computer Vision-Driven Mitigation of Marasmius Palmivorus in Oil Palm Plantations

Abstract

Marasmius palmivorus is an emerging fungal threat associated with progressive canopy degradation and yield loss in oil palm plantations. Conventional mitigation practices remain largely reactive due to delayed visual recognition, limited field coverage, and heterogeneous environmental conditions. This paper presents a deployment-oriented experimental study investigating the robustness of computer vision (CV)-based detection under real plantation variability. Field data were collected using binocular-assisted video acquisition to capture illumination heterogeneity, canopy occlusion, and background clutter representative of operational conditions. A YOLOv12-based detection framework was implemented and enhanced using illumination normalization strategies, including Single-Scale Retinex (SSR) and an Improved Single-Scale Retinex (ISSR) formulation. Comparative evaluation using mAP@50–95 and stability analysis demonstrates that ISSR improves detection consistency under heterogeneous lighting without increasing model complexity. Beyond algorithmic accuracy, this study establishes an explicit mapping between CV outputs and actionable mitigation decisions and proposes a scalable deployment architecture integrating edge inference, centralized dashboards, and human-in-the-loop verification. The findings highlight the importance of robustness-aware evaluation and deploymentconscious design in translating computer vision models into reliable decision-support systems for sustainable oil palm disease management.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint