Automated Coconut Maturity Detection and Timeline Prediction using YOLOv8-based Computer Vision

The proper determination of coconut maturity is difficult in enhancing agricultural output and streamlining harvesting time but the conventional techniques involve a manual process overly relying on subjective judgment. To overcome such limitations, this research presents a new, fully automated system CocoVision-Edge, a maturity detection system and dynamic timeline prediction of coconuts to be used in precision agriculture. The framework consists of a lightweight, attention-enhanced YOLOv8 deep learning framework, optimized to run on Unmanned Aerial Vehicles (UAVs) and edge devices with extremely low latency. The system powerfully recognizes coconuts in three major maturity phases, namely, premature, potential and mature with a minimum mean Average Precision (mAP50) of 76.1 even in the presence of challenging canopy lighting and challenging backgrounds. To go beyond conventional fixed classification. This research presents two significant advancements: a dynamic harvest schedule prediction framework that approximates the time left to harvest, precisely ready to lay the foundation of weather-sensitive scheduling, and the incorporation of Explainable AI (XAI) with the help of visual saliency maps that emphasize the phenotypic attributes that the model relies on to make its choices. Being implemented as a decentralized, edge-friendly app, CocoVision-Edge requires no cloud infrastructure whatsoever, providing real-time, explainable and predictive insights at the field to farmers immediately.