Case Study: Sterblue achieves 13% more accurate defect detection and 50% reduction in manual inspection time with Aquarium Learning

Optimizing Infrastructure Inspection With Machine Learning

Sterblue builds software to manage infrastructure inspections by analyzing imagery from drones, helicopters, satellites, and smartphones to detect defects in power lines, wind turbines, and other assets. Facing the need for higher detection accuracy, greater data throughput, and reduced human inspection across varied datasets, Sterblue partnered with Aquarium Learning and used Aquarium Learning’s machine learning data management platform to inspect label quality and optimize their ML pipeline.

Aquarium Learning’s platform let Sterblue visualize labeled datasets, analyze geographic data distribution, pinpoint where model–label disagreements indicated labeling errors, and integrate issue-tracking with their in‑house labeling and training pipelines. Using Aquarium Learning, Sterblue retrained a cleaner dataset and deployed a model 13% more accurate than the prior version, cut manual expert inspection time by 50%, scaled defect detection to several hundred miles of power lines, and saved significant engineering and labeling effort.

Sterblue