An FL-based underwater lidar image classification system ensuring an adversarially robust functioning environment. It is mainly applicable to lightweight underwater navigation systems with limited computational capabilities. It utilizes a set of underwater lightweight devices as data generators and training locations, and a separate on-land base station acts as the aggregator. In addition, the onsite training of the model leads lesser bandwidth for data transfer.

Breakthrough/World class work in the project:

The purpose system introduces an adversarially robust LIDAR image based underwater object detection model developed on FL schema.

Impact on the Country:

The system ensures safer and much cost effective experience for local fisherfolks during their fishing adventures operating on small scale fishing vessels.

Novel feature in the project:

A low cost system for small scale fishing vessels for profitable experience

The system can efficiently detect and identify entities in locally generated datasets

Sectors & Applications:

Local fisherfolk and smaller fishing vessels

Small shipping vessels where radar mechanism are not feasible for fisheries management, underwater environment monitoring