Every spring, river herring embark on a critical migration from coastal waters to freshwater spawning grounds in Massachusetts. Traditionally, monitoring these populations has relied on labor-intensive visual counts by volunteers—a method limited by daylight and human stamina. Researchers from MIT Sea Grant and CSAIL have introduced a sophisticated alternative: a neural network-based system capable of continuous, automated fish monitoring via underwater video.

The team developed an end-to-end pipeline that handles everything from raw video capture to object detection and species classification. By training models on diverse datasets reflecting various water conditions and lighting, the system successfully identified and tracked thousands of fish. Crucially, the AI revealed biological insights humans had missed, such as peak upstream movement at dawn and predominantly nocturnal downstream migration, which helps fish avoid predators under the cover of darkness.

While the technology offers unprecedented scale and precision, the researchers emphasize that it is designed to augment, not replace, citizen science. Volunteers remain vital for maintaining equipment and verifying model outputs. This collaborative framework between human observers and automated visual analysis sets a new standard for cost-effective, high-resolution environmental conservation and fisheries management across diverse aquatic ecosystems.

Every spring, river herring embark on a critical migration from coastal waters to freshwater spawning grounds in Massachusetts. Traditionally, monitoring these populations has relied on labor-intensive visual counts by volunteers—a method limited by daylight and human stamina. Researchers from MIT Sea Grant and CSAIL have introduced a sophisticated alternative: a neural network-based system capable of continuous, automated fish monitoring via underwater video.

The team developed an end-to-end pipeline that handles everything from raw video capture to object detection and species classification. By training models on diverse datasets reflecting various water conditions and lighting, the system successfully identified and tracked thousands of fish. Crucially, the AI revealed biological insights humans had missed, such as peak upstream movement at dawn and predominantly nocturnal downstream migration, which helps fish avoid predators under the cover of darkness.

While the technology offers unprecedented scale and precision, the researchers emphasize that it is designed to augment, not replace, citizen science. Volunteers remain vital for maintaining equipment and verifying model outputs. This collaborative framework between human observers and automated visual analysis sets a new standard for cost-effective, high-resolution environmental conservation and fisheries management across diverse aquatic ecosystems.