ship detection

Real-Time Ship Detection Model Improves Maritime Intelligence Through Integrated Data

Real-time ship detection is paramount for maritime safety and intelligent navigation, yet existing models struggle with accuracy and performance under challenging conditions.

September 22, 20265 min readFrontiers in Marine Science | New and Recent Articles

Real-Time Ship Detection Model Improves Maritime Intelligence Through Integrated Data

Our take

The increasing frequency of maritime incidents underscores the vital need for robust and reliable ship detection technology. Recent events, such as the tragic sinking of a tugboat off Busan [1 Dead, 6 Missing After Tugboat Sinks While Towing Container Ship Off Busan] and the collision of two merchant ships off India's Gujarat coast [21 Rescued After Two Merchant Ships Collide Off India’s Gujarat Coast], highlight the potential for devastating consequences when maritime situational awareness is compromised. The challenges of locating survivors, coordinating rescue efforts, and preventing further accidents are significantly amplified by limited visibility and the vastness of the ocean. This new research, detailing the SGR-YOLO model, represents a tangible step towards improving the accuracy and speed of ship detection, ultimately contributing to safer and more efficient maritime operations. The reliance on real-time data and advanced algorithms is increasingly crucial in an environment where even brief delays can have profound implications, as tragically illustrated by the experience of ferry survivors waiting for rescue [Indonesia’s Ferry Survivor Saw Ship Overturn In Minutes, Then Waited 10 Hours For Rescue].

The SGR-YOLO model’s improvements—particularly its incorporation of Sim-BiFPN, GhostConv, and RepViT—address critical limitations in existing ship detection systems. The focus on lightweight feature fusion and reduced computational complexity is especially noteworthy. Traditional methods often struggle to balance accuracy with real-time performance, a crucial consideration for practical deployment in resource-constrained monitoring systems. The ability to maintain high detection precision even under adverse conditions like fog, occlusion, and long distances demonstrates the model’s potential for enhancing maritime safety across a range of operational scenarios. The validated reduction in parameters—5.2M on the SeaShips dataset and 2.9M on the McShips dataset—suggests a significant efficiency gain without compromising performance, facilitating wider adoption and integration into existing infrastructure. This exemplifies the kind of iterative, empirical refinement necessary for advancing the field of ocean intelligence.

Beyond the immediate benefits for maritime safety, the SGR-YOLO model’s advancements have broader implications for ocean stewardship and climate monitoring. Accurate and timely ship detection is fundamental for tracking vessel movements, monitoring compliance with environmental regulations, and assessing the impact of maritime activities on marine ecosystems. Integrated data ecosystems, such as the one this model contributes to, are essential for generating comprehensive ocean intelligence, allowing for more informed decision-making regarding resource management and climate mitigation strategies. The longitudinal data generated through improved detection capabilities can be leveraged to identify trends, predict potential risks, and develop targeted interventions to protect vulnerable marine environments. The shift towards more sophisticated, calibrated models underscores the growing recognition of data-driven approaches in addressing complex ocean challenges.

Looking ahead, the continued development of real-time ship detection models like SGR-YOLO will be critical for supporting the growing demands of autonomous maritime systems and the increasing need for comprehensive ocean monitoring. The integration of these models with satellite-based observation platforms and other sensor networks holds significant promise for creating a truly integrated ocean intelligence system. A key question remains: how can we ensure the equitable distribution and accessibility of these advanced technologies, particularly for nations and communities most vulnerable to the impacts of climate change and maritime hazards? The collaborative nature of this research, and the open-source availability of similar models, suggests a positive trajectory, but continued investment in global data sharing and capacity building will be essential to realize the full potential of this transformative technology.

Ship detection is a critical technology for maritime safety supervision, collision avoidance, and intelligent navigation. However, existing methods face significant challenges, including large scale variations of ship targets, severe background interference under complex weather conditions (e.g., fog, low light), and the contradiction between model accuracy and real-time performance. To solve these problems, this study proposes an improved model named SGR-YOLO based on YOLOv8n. First, Sim-BiFPN is introduced in the neck as a lightweight cross-scale feature-fusion structure, with SimAM embedded within its fusion nodes to adaptively emphasize ship-related features. Second, the standard convolution in the C2f module is replaced with GhostConv to construct the GhostC2f module, which achieves a significant reduction in computational complexity without sacrificing detection accuracy. Third, RepViT is placed in the neck after the final GhostC2f block and immediately before the small-object detection branch, strengthening global context modeling for small and distant ships. Extensive experiments are conducted on two public datasets (SeaShips and McShips). The results demonstrate that the SGR-YOLO model achieves a detection precision of 98.7% on the SeaShips dataset with a 5.2M reduction in parameters, and 93.3% on the McShips dataset with a 2.9M parameter reduction. Furthermore, this model maintains high-robust detection performance under various maritime conditions, including foggy weather, target occlusion, and long-distance scenarios. This study provides a feasible solution for real-time and high-precision ship detection in resource-constrained maritime monitoring systems.

Read the original at Frontiers in Marine Science | New and Recent Articles