HAMSCNN : a Hybrid Attention Multi-Scale CNN for accurate ship detection in maritime surveillance

Abstract

Maritime surveillance is a critical component of global security, trade monitoring, and environmental protection, with ship detection playing a key role in preventing illegal activities and ensuring safe navigation. Satellite imagery, combined with deep learning techniques, has emerged as a powerful solution for real-time ship detection, offering enhanced accuracy and scalability over traditional methods. For the study, a novel deep learning model, Hybrid Attention Multi-Scale CNN (HAMSCNN), is proposed to accurately classify satellite images into Ship and No-Ship categories. The model integrates channel attention, spatial attention, and multi-scale convolution blocks to enhance feature extraction, emphasize critical spatial and channel-wise information, and improve detection across varying ship sizes and oceanic conditions. The dataset comprises 4,000 satellite images, with 1,000 Ship images and 3,000 No-Ship images, addressing the challenge of class imbalance through data augmentation techniques such as flipping, zooming, and rotation. The HAMSCNN model incorporates batch normalization, dropout layers, and max-pooling for improved regularization and stability. Extensive experiments demonstrate that the proposed model significantly outperforms existing state-of-the-art architectures, achieving a classification accuracy of 99%, with precision, recall, and F1-scores all reaching 0.99. The confusion matrix highlights the model’s reliability, with minimal misclassifications (6 false positives and 4 false negatives), reinforcing its suitability for real-world deployment. The results underscore the potential of HAMSCNN in advancing autonomous ship detection, enabling efficient, cost-effective, and scalable solutions for maritime monitoring. Future work will explore the integration of Synthetic Aperture Radar (SAR) data to enhance robustness under adverse weather conditions and expand the model’s applicability to multi-class maritime object detection.