Monday 31 March 2025
Underwater robots are getting a boost in their ability to detect and identify objects on the seafloor, thanks to a new dataset of underwater images that can be used to train machine learning models. The Common Objects Underwater (COU) dataset contains over 10,000 images of man-made objects like diving gear, marine debris, and even underwater vehicles, all taken in various aquatic environments.
The COU dataset is the result of a collaboration between researchers from several institutions who wanted to create a more comprehensive and diverse collection of underwater images. The goal was to help improve the accuracy and efficiency of underwater object detection systems, which are crucial for tasks like search and rescue operations, environmental monitoring, and marine conservation.
Traditionally, underwater robots have relied on human operators to manually identify objects in real-time, but this approach has limitations. For one, it’s time-consuming and labor-intensive, and it can be difficult to detect small or hard-to-reach objects. Machine learning models, on the other hand, can quickly process large amounts of data and learn to recognize patterns, making them ideal for underwater object detection.
The COU dataset is specifically designed to help train these machine learning models. The images were captured in a variety of environments, including pools, lakes, and oceans, and feature objects from different angles and lighting conditions. This diversity will allow the models to learn to recognize objects regardless of their orientation or the surrounding environment.
The researchers used three state-of-the-art object detection models – YOLOv9-C, Mask R-CNN, and Mask2Former – to test the COU dataset’s effectiveness. They found that all three models performed well on the dataset, with the highest accuracy achieved by Mask2Former. The results show that the COU dataset can be used to train accurate object detection models, which could ultimately lead to more efficient and effective underwater operations.
One of the key challenges in developing underwater object detection systems is the limited availability of high-quality training data. The COU dataset addresses this issue by providing a large collection of images that can be used to train machine learning models. This will not only improve the accuracy of underwater object detection but also enable researchers to develop more sophisticated algorithms and models.
The impact of the COU dataset could be significant. For example, it could help search and rescue teams quickly locate missing divers or debris in the aftermath of a disaster. It could also aid marine conservation efforts by enabling researchers to monitor and track marine life more effectively.
Cite this article: “Underwater Object Detection Boosted with New Dataset”, The Science Archive, 2025.
Underwater Robots, Object Detection, Machine Learning, Underwater Images, Cou Dataset, Marine Debris, Diving Gear, Underwater Vehicles, Search And Rescue, Environmental Monitoring
