In the vast, open seas, spotting small vessels like patrol boats can be a challenge, especially when they’re dwarfed by the surrounding environment. That’s where the work of Zhaohua Liu, a researcher at the School of Physics and Electronic at Yantai University in China, comes into play. Liu and his team have developed a new algorithm to improve the detection of small maritime targets, and their findings were recently published in the journal ‘Sensors’ (translated from the original Chinese title).
The team’s work focuses on improving the You Only Look Once version 11n (YOLOv11n) algorithm, a popular deep learning model used for object detection. The improved algorithm, dubbed the maritime small target image detection algorithm, aims to tackle the issues of small size, insufficient feature information, and high missed detection rates that plague current systems.
So, what’s new here? Liu and his team introduced a few key improvements to the YOLOv11n model. First, they added a module called BIE (Bidirectional Interaction Enhancement) to the feature fusion stage. This module helps to progressively fuse features from both infrared and visible light images, creating independent branches for key features of small maritime targets.
Next, they incorporated a module called RepViTBlock into the backbone network. This module, combined with the existing C3k2 module of YOLOv11n, enhances the model’s ability to extract local features of small targets. Liu explains, “Through the lightweight attention mechanism and multi-branch convolution structure, this addresses the insufficient capture of tiny target features by the C3k2 module.”
Finally, the team embedded a ConvAttn module at the end of the backbone network. This module uses dynamic small-kernel convolution to adaptively extract the contour features of small targets, reducing the missed detection rate while keeping the model lightweight.
The results speak for themselves. When tested on two datasets—one collected by the team (IVships) and one public dataset (SeaShips)—the improved algorithm showed significant improvements. It increased the mean average precision ([email protected]) by 1.9% and 1.7%, respectively, and the average precision by 2.2% and 2.4%, respectively, compared to the original model.
So, what does this mean for the maritime industry? Improved small target detection can enhance maritime safety, security, and efficiency. For instance, it can help in monitoring small vessels for search and rescue operations, detecting illegal fishing activities, or even spotting small debris that could pose a hazard to larger ships.
Moreover, the improved algorithm could find applications in autonomous shipping, where accurate and reliable object detection is crucial for safe navigation. As Liu points out, “The improved algorithm significantly improves the model’s small target detection capabilities.”
In conclusion, Liu’s work represents a significant step forward in maritime small target detection. By improving the YOLOv11n algorithm, the team has opened up new opportunities for enhancing maritime safety and security. As the maritime industry continues to evolve, such advancements will play a pivotal role in shaping its future.