Unmanned surface vehicle with deep learning-based obstacle avoidance for water quality monitoring

This study presents the design, implementation, and critical evaluation of an unmanned surface vehicle (USV) for real-time water quality monitoring in harbour environments equipped with an AI-based vision system for obstacle avoidance. The primary objective was to develop a mobile platform capable of gathering multi-parameter water data (pH, temperature, turbidity, total dissolved solids) at 30-second intervals and transmitting it remotely while autonomously navigating and avoiding collisions. Key findings indicate that the USV successfully integrates environmental sensing with deep learning-based obstacle detection, achieving a maximum obstacle detection range of 3.4m in optimal daylight and meeting water sampling frequency requirements. Strengths of the system include a stable catamaran hull design that exceeded payload capacity targets (15 kg carried vs 10 kg target) and an innovative vision approach using semantic segmentation to distinguish water and sky from obstacles. The research’s main contributions lie in combining reliable water monitoring with AI-driven navigation on a low-cost platform. Overall, the USV demonstrates the viability of combining deep learning vision with environmental monitoring, but further refinement is recommended in obstacle avoidance algorithms, sensor calibration, and extended field testing to ensure robust operation under diverse real-world conditions.