What question did this study set out to answer?

The aim is to develop a robust underwater monitoring system that overcomes challenges posed by marine pollution and climate change.

June 7, 2026Open Access

Advanced underwater monitoring for environmental protection and resource exploration using deep learning-based hybrid sensor network

Key Points

The aim is to develop a robust underwater monitoring system that overcomes challenges posed by marine pollution and climate change.
Proposed a Hybrid Underwater Wireless Sensor Network integrating acoustic and optical sensing with deep learning.
Utilized mean-shift tracking for object localization and arithmetic coding for data compression.
Evaluated the framework on a dataset of 7,212 images using hierarchical feature extraction and anomaly detection techniques.
Achieved 98.0% Average Precision for object tracking and 99.1% anomaly detection accuracy.
Demonstrated overall classification accuracies of 98.63% during training and 99.30% on a test set.
Confirmed significant superiority over baseline models with p < 0.001 using a paired t-test.

Abstract

Abstract The escalating impacts of marine pollution and climate change demand reliable underwater monitoring systems capable of operating under harsh subsea conditions, including high pressure, low visibility, and constrained communication bandwidth. This paper proposes a Hybrid Underwater Wireless Sensor Network (UHWSN) that integrates acoustic and optical sensing with advanced deep learning to enable efficient and accurate underwater monitoring. At the sensor layer, mean-shift tracking is employed for robust object localization, while arithmetic coding performs lossless compression, reducing transmitted data volume by up to 60% without information loss. At the cluster-head layer, Convolutional Neural Networks (CNNs) conduct hierarchical feature extraction, and Deep Belief Networks (DBNs) provide probabilistic object classification and anomaly detection. The framework was evaluated using a video-group-aware stratified 80/20 split on a combined dataset of 7,212 unique images from the TrashCan and J-EDI underwater benchmarks. Experimental results demonstrate strong performance, achieving 98.0% Average Precision (AP) for object tracking, 99.1% anomaly detection accuracy, and overall classification accuracies of 98.63% during training and 99.30% on a strictly held-out test set. Regression analysis further confirms high predictive reliability, yielding a perfect linear correlation ( R = 1) between predicted and ground-truth values with uniformly distributed residuals. Statistical validation using a paired t-test ( p < 0.001) demonstrates significant superiority over baseline and ablation models. Compared with state-of-the-art approaches, the proposed framework consumes 1.25 J per processed image in simulation, representing a 35–55% reduction relative to ablations and corresponding to 30–40% communication overhead savings. These results highlight the effectiveness of the proposed UHWSN framework for energy-efficient, high-precision underwater monitoring in resource-constrained environments.

Ask AI

Helpful

Bookmark

View Full Paper