A Novel Hybrid Deep Reinforcement Learning Approach for Dynamic Biofouling Detection and Structural Health Monitoring in Marine Environments

The complex, dynamic and low-visibility conditions of the marine environment make biofouling detection and structural health monitoring extremely difficult. More conventional techniques and methodologies are grounded in manual inspection, making them labor and error intensive. This paper proposed a novel hybrid Deep Reinforcement Learning (DRL) framework that integrates multi level Convolutional Neural Networks (CNNs) for fine grained biofouling classification and Graph Neural Networks (GNNs) for robust structural integrity assessment. The multi-objective reward function allows to promote high accuracy in detecting biofouling (92.3%), low mean squared error in predicting the structure (MSE of 0.021), as well as energy efficient navigation. Under turbulent water conditions, a 32% decrease in inspection time and an 18.7% increase in detection precision is shown for our model against traditional techniques. The developed system represents the first integrated DRL approach for real-time dual-task DRT and DNL monitoring and it paves the way for the deployment of scalable autonomous underwater detection systems.