A Dual-Stream Deep Learning Framework for Proactive Maritime Safety and Yield Optimization (IMSAS)

DOI

https://www.doi.org/10.59256/ijire.20260702025

Published In

Volume 7, Issue 2 (2026)

Published Online

March-April 2026

Pages

203-208

Article Type

Original Article

Check for updates

The unpredictable nature of the sea continues to pose severe safety and economic challenges for coastal fishing communities worldwide. Existing maritime safety systems operate reactively, alerting fishermen only when danger is imminent or boundaries are crossed, thereby offering no capacity for proactive risk avoidance or resource optimization. This paper presents the Intelligent Maritime Safety and Assistance System (IMSAS), a dual-objective, data-driven framework that integrates predictive hazard forecasting with optimal fishing zone identification. IMSAS employs a Long Short-Term Memory/Gated Recurrent Unit (LSTM/GRU) deep learning model for time-series sea-state forecasting — achieving 12-hour prediction lead times with 93.5% accuracy and an F1-score of 0.91 — alongside a Convolutional Neural Network (CNN) for spatial extraction of Potential Fishing Zones (PFZ) from satellite imagery (SST and Chlorophyll-a), attaining 96.13% pixel accuracy and a correlation of 0.82 with historical catch logs. A fusion decision layer combines both model outputs into unified Red/Green Zone map overlays and automated pre-trip alerts, delivered via a cloud-native, scalable infrastructure. Experimental results confirm the system's feasibility and significant promise for improving safety, operational efficiency, and ecological sustainability in the fishing industry.