Deep learning-enabled hybrid sensor network for intelligent underwater monitoring
Résumé
Abstract The increasing prevalence of marine pollution and the escalating impacts of climate change have intensified the need for advanced underwater monitoring systems capable of operating reliably in harsh subsea environments. However, underwater environmental monitoring remains a challenging task due to factors such as high hydrostatic pressure, limited visibility, dynamic water conditions, and the inherent constraints of underwater communication channels. To address these challenges, this study proposes a Hybrid Underwater Wireless Sensor Network (UHWSN) that integrates acoustic and optical sensing modalities with advanced deep learning techniques to enable efficient underwater debris detection, tracking, and classification. The proposed framework combines mean-shift-based object tracking, lossless arithmetic coding, Convolutional Neural Networks (CNNs), and Deep Belief Networks (DBNs) within a hierarchical processing architecture. At the sensor-node level, mean-shift tracking is employed for region-of-interest (ROI) localization, while arithmetic coding provides efficient lossless data compression, achieving a data-volume reduction of approximately 55–60% without compromising image fidelity. At the cluster-head level, CNN-based feature extraction and DBN-based classification facilitate robust high-level inference and anomaly detection under challenging underwater conditions. The performance of the proposed UHWSN framework was evaluated using a combined TrashCan and J-EDI dataset comprising 7,212 images, augmented to simulate realistic underwater degradations, including turbidity, low-light conditions, and color distortions. Experimental results demonstrate a tracking Average Precision (AP) of 98.0% and a classification accuracy of 99.30% on the test set. These results compare favorably with recent state-of-the-art approaches evaluated on similar underwater debris detection benchmarks, which typically report detection performance in the range of 65–90% mAP under comparable conditions. Furthermore, the proposed framework achieves enhanced energy efficiency and reduces communication overhead by approximately 30–40% relative to conventional underwater monitoring architectures, thereby extending network lifetime and improving operational scalability in resource-constrained environments.
Citer ce document
Accès au document
Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter
Voir l'article sur le site de la revueAuteur(s)
Statistiques
Consultations : 1
Téléchargements : 0