Leveraging Large Language Models for IoT Applications: A Maritime Image Dataset Perspective

This is the authors' abstract. We don't add key points for this paper.

The increasing reliance on advanced image processing techniques in maritime surveillance, autonomous navigation, and remote sensing underscores the need for robust, efficient, and scalable AI-driven solutions. However, the inherent challenges of dynamic oceanic environments—such as fluctuating lighting conditions, occlusions, and computational constraints of Internet of Things (IoT) devices—limit the effectiveness of traditional deep learning approaches. This study presents a novel hybrid framework that integrates Large Language Models (LLMs) with deep learning-based image analysis to enhance maritime IoT applications. By leveraging multimodal data fusion, adaptive preprocessing, and ensemble learning, the proposed approach significantly improves detection accuracy, robustness, and real-time decision-making capabilities in resource-constrained maritime environments. Extensive experimental evaluations on benchmark datasets, including the Singapore Maritime Dataset and the Maritime Object Detection System (MODS), demonstrate the efficacy of the proposed framework. The results indicate a classification accuracy of 94.3%, with a 24% reduction in computational overhead compared to conventional deep learning methods. Furthermore, the study explores key advancements in AI-driven maritime monitoring, anomaly detection, and autonomous vessel navigation, highlighting the transformative potential of LLM-enhanced image processing for next-generation maritime IoT ecosystems. Additionally, this work outlines future research directions, focusing on optimizing computational efficiency, enhancing cybersecurity measures to protect against adversarial threats, and integrating emerging technologies such as federated learning and edge AI to further improve maritime situational awareness and operational resilience.