Enhanced Vessel Detection for Maritime Surveillance Using Hyperparameter-Tuned Deep Learning on SAR Images

Abstract

In maritime surveillance, where ensuring the safety of shipping routes and detecting potential threats is paramount, the need for efficient vessel detection models is critical. Unauthorized vessels pose a significant threat to maritime security, competing for essential resources such as shipping lanes and port access. Traditional methods of vessel detection, such as manual monitoring or blanket radar scans, are time-consuming, labor-intensive, and often result in overuse of resources, leading to operational inefficiency and potential security breaches. This paper presents a Hyperparameter-Tuned Deep Learning model for Vessel Detection and Classification (HPTDL-VDAC) suitable for Maritime Surveillance applications. The proposed HPTDL-VDAC system integrates advanced techniques from computer vision and deep learning to accurately identify and classify vessels in Synthetic Aperture Radar (SAR) images. The workflow begins with pre-processing steps aimed at enhancing image quality and reducing noise. Specifically, a Gaussian Filter (GF) is employed to effectively remove noise from input images, followed by resizing to standard dimensions for subsequent analysis.
For object detection and classification, the RetinaNet model is employed. RetinaNet's innovative architecture, featuring a focal loss mechanism, enables robust detection of vessel instances amidst varying backgrounds and sea conditions. Notably, the hyperparameters of the RetinaNet model are fine-tuned using the ADAM optimizer, optimizing its performance for the specific task of vessel detection in maritime surveillance scenarios. A thorough simulation analysis of the HPTDL-VDAC technique was conducted using a benchmark dataset. Experimental results demonstrate the effectiveness of the proposed system in accurately detecting vessels in various maritime environments. This shows that it exhibits improved results compared to recent approaches on various metrics.