Underwater AI detects marine life faster on power-limited vehicles

Tests found the system used less computing power and ran faster than a larger comparison model, with only small decreases in reported accuracy.

Editorial illustration — not from the study.

Researchers developed LMOD-YOLO, a smaller AI system designed to detect marine organisms in difficult underwater images while using limited computing resources. On the CUDD dataset, it used fewer parameters and less computation than YOLOv8m while maintaining similar reported accuracy.

When deployed on an NVIDIA Jetson AGX Orin, the system had a mean inference latency of 2.90 milliseconds and an energy efficiency of 11.76 frames per watt. The authors say this combination could support timely monitoring on battery-powered autonomous underwater vehicles.

What the model achieved

LMOD-YOLO reduced computational cost by 51.1% and the parameter count by 59.7% compared with YOLOv8m. Its inference speed increased by 36.9%, while mean average precision fell by 0.006 at the 0.5 threshold and by 0.012 across thresholds from 0.5 to 0.95. On an NVIDIA Jetson AGX Orin, the model reached a mean inference latency of 2.90 milliseconds and an energy efficiency of 11.76 frames per watt. The model combines changes intended to reduce repeated computation, improve the merging of image features at different scales, and handle classification and position estimates more efficiently during deployment.

Why lower power matters

Marine-organism detection can help with underwater ecological monitoring, protection of endangered species and autonomous underwater vehicle operations. A system that runs quickly with lower computing demands could help battery-powered vehicles respond in a timely way while reducing their energy burden, although the abstract reports testing on a dataset and an embedded computer rather than a live vehicle mission.

Evidence and caveats

The evidence comes from experiments on the CUDD dataset and a deployment test on an NVIDIA Jetson AGX Orin, with performance compared with YOLOv8m. The abstract does not report results from a live underwater vehicle or explain how well the system transfers to other waters, species, cameras or hardware. The reported accuracy measures decreased slightly compared with YOLOv8m.

Reliable AI partnerships depend on people, not just stronger models

A survey of human-AI collaboration argues that more capable AI systems do not automatically produce reliable partnerships. It identifies human-centered design, governance and attention to high-stakes uses as central to making these systems trustworthy and beneficial.

9 hours ago·6 views

AI & Computing

RNA splicing tools can follow irrelevant clues and miss RNA’s shape

Deep-learning tools for predicting RNA splicing can rely on genomic clues unrelated to splicing and miss important effects of RNA structure, a study finds. These weaknesses were linked to systematic errors, especially when the tools evaluated sequences that differed from those used for training.

20 hours ago·5 views

AI & Computing

Better-trained AI finds more flaws in unfamiliar software projects

AI vulnerability detectors often perform less well on software projects outside their training data. A study finds that cleaner, more varied training data and an encoder-based design improved detection, including a reported 6.8% gain in recall on the BigVul benchmark.

2 days ago·34 views