China opens world’s largest ship data set that could be used to train drones

A consortium of Chinese military and academic institutions has publicly released a significant new resource for training artificial intelligence in maritime warfare. The Dual-Modal Ship Detection (DMSD) dataset, developed by researchers from the Naval Aeronautical University, Harbin Engineering University, and the Chinese Academy of Sciences, is billed as the world's first publicly available dataset pairing visible light and infrared imagery of ships. It contains more than 2,000 matched image pairs and nearly 20,000 annotated ship instances, captured under various sea states and conditions. The dataset is specifically engineered to address the unique challenges of maritime target recognition, where factors like glare, weather, long distances, and cluttered backgrounds severely degrade the accuracy of AI vision systems.

This release directly targets a critical military capability gap: reliably identifying and tracking naval targets in contested environments where radar may be jammed or degraded. The need was highlighted by a recent incident in the Strait of Hormuz, where Iran claimed a successful strike on a U.S. aircraft carrier, but U.S. officials stated the attacking drones or missiles failed to properly identify and engage the target. By providing paired sensor data, the DMSD dataset allows AI models to be trained for multi-spectral fusion, enabling autonomous systems like drones or missiles to maintain target recognition at night, through smoke, or in other visually obscured scenarios where a single sensor type would fail.