Detection and Classification of Mangoes using Machine Learning on Various Edge AI devices in Real-Time through Drone

Accurate and efficient detection of mangoes on trees plays a vital role across every stage of the agricultural value chain. Traditional manual methods are time-consuming, subjective, and prone to error. This paper explores the potential of machine learning (ML) for automated mango detection in real-time on edge AI devices. We investigate the use of YOLOv5 and YOLOv8 to accurately classify mangoes into three categories - ripe, unripe and infected ones. These models are trained on a custom dataset collected from various sources. The trained models are then deployed on various edge AI devices like Raspberry Pi 4B, NVIDIA Jetson Nano and NVIDIA Jetson Orin. Live video captured by a UAV drone is fed as an input to the edge AI device for real-time inferencing. The experimental results demonstrate the accuracy of YOLOv8 to be 94.4% and YOLOv5 be of 93.1%. These models offer a good balance between accuracy and speed, making them suitable for resource-constrained edge environments. By leveraging machine learning and edge AI, this work aims to demonstrate a novel approach for automated, efficient, and data-driven mango detection, which can aid in ensuring better quality produce.