A Machine Learning Framework for Dynamic Biofouling Detection and Structural Health Monitoring in Marine Environments

Currently, the autonomous underwater vehicle (AUV) technology is undergoing a rapid evolutionary transition due to the increasing demand for exploration and operational capabilities. This paper introduces a new framework integrating deep reinforcement learning (DRL) with hybrid convolutional neural networks (CNNs) to empower AUVs with autonomous biofouling detection and structural health monitoring. The proposed system can detect different types of biofouling and assess the health of the underwater structure simultaneously. A multi-objective reward function is introduced to promote high accuracy in both the detection of biofouling (92.3%) and the prediction of the structure (mean squared error of 0.021), while maintaining low energy consumption. The system shows a 32% reduction in inspection time and 18.7% improvement in detection accuracy compared to traditional techniques under turbulent water conditions.