In the ever-evolving landscape of maritime technology, a groundbreaking development has emerged from the College of Fishery at Ocean University of China, spearheaded by lead author Lei Zhang. Published in the Journal of Marine Science and Engineering, the research introduces a novel algorithm designed to revolutionize ship plate detection, a critical component of intelligent maritime supervision.
The algorithm, dubbed RT-DETR-HPA, is a multi-module collaborative detection system built upon an enhanced RT-DETR framework. It’s designed to tackle the complex challenges of ship plate detection, such as small target sizes, extreme aspect ratios, dense arrangements, and multi-angle rotations. “Our goal was to create a system that could accurately detect ship plates in even the most challenging maritime scenarios,” Zhang explained.
At the heart of RT-DETR-HPA are three core components. The first is an improved High-Frequency Enhanced Residual Block (HFERB) embedded in the backbone, which strengthens multi-scale high-frequency feature fusion. This is complemented by deformable convolution to handle occlusion and deformation. The second component is a Pinwheel-shaped Convolution (PConv) module, which employs multi-directional convolution kernels to achieve rotation-adaptive local detail extraction. This allows for accurate capture of plate edges and character features. The third component is an Adaptive Sparse Self-Attention (ASSA) mechanism incorporated into the encoder. This automatically focuses on key regions while suppressing complex background interference, thereby enhancing feature discriminability.
The results of the research are impressive. Comparative experiments conducted on a self-constructed dataset of 20,000 ship plate images showed that RT-DETR-HPA achieved a 3.36% improvement in mAP@50, reaching 97.12%, and a 3.23% increase in recall, reaching 94.88%. Moreover, it maintained real-time detection speed at 40.1 FPS. Compared to mainstream object detection models like the YOLO series and Faster R-CNN, RT-DETR-HPA demonstrated significant advantages in high-precision localization, adaptability to complex scenarios, and real-time performance.
The commercial impacts and opportunities for the maritime sector are substantial. Accurate and real-time ship plate detection is crucial for a variety of applications, including port security, vessel traffic management, and maritime law enforcement. The RT-DETR-HPA algorithm can significantly enhance the efficiency and accuracy of these operations, leading to improved safety and security in maritime environments.
Moreover, the algorithm’s ability to reduce missed and false detections caused by low resolution, poor lighting, and dense occlusion makes it a robust and high-accuracy solution for intelligent ship supervision. As Zhang noted, “Our algorithm provides a reliable tool for maritime professionals to navigate the complexities of ship plate detection.”
Looking ahead, the research team plans to focus on lightweight model design and dynamic resolution adaptation to enhance the algorithm’s applicability on mobile maritime surveillance platforms. This ongoing innovation promises to further advance the capabilities of maritime technology, opening up new possibilities for the industry.
In the realm of maritime technology, the work of Lei Zhang and his team represents a significant leap forward. Their research, published in the Journal of Marine Science and Engineering, offers a powerful tool for enhancing maritime supervision and security, underscoring the potential of advanced algorithms in addressing complex maritime challenges.