A Computer Vision-Based Autonomous System for Ripe Strawberry Detection and Navigation

Abstract

This work presents an autonomous robotic system designed for ripe strawberry detection and navigation in agricultural environments using cost-efficient hardware and optimized algorithms. The system integrates a Jetson Nano for real-time image processing, employing LAB color space transformation, morphological filtering, and a lightweight MobileNet SSD model trained on the StrawDI dataset to classify strawberries by ripeness. Autonomous navigation is achieved via a Pixhawk flight controller that tracks GPS waypoints and adjusts trajectories dynamically using ultrasonic sensors for obstacle detection. Field experiments in a controlled strawberry field setup demonstrated robust performance: the vision system achieved 86.21% precision and 89.29% recall. Navigation tests showed 95% accuracy in following a 5-meter path, with successful obstacle avoidance. The results validate the feasibility of deploying low-cost, energy-efficient embedded systems for agricultural automation, addressing scalability challenges in resource-limited settings.