srimanta-mandal/Multi-Fish-Tracking-with-Underwater-Image-Enhancement

# Multi-Fish-Tracking-with-Underwater-Image-Enhancement-by-Deep-Network-in Marine-Ecosystems Tracking of marine life is an important part of the study and analysis of migration patterns, movements, and population growth of underwater animals. Fish-tracking deep learning networks have been under research and development for quite some time and have been able to produce satisfactory results. We propose an end-to-end deep learning framework to track fishes in unconstrained marine environment. The novel and a key component of this network is image enhancement based Siamese architecture that measures appearance similarity. The image enhancement is performed by the proposed enhancement module, which consists of convolutional layers along with a squeeze and excitation block. The module is pre-trained using degraded and clean image pairs to deal with the underwater degradation. The enhanced feature is utilized by the Siamese framework to produce an appearance similarity score, which is further complemented by prediction scores that consider the fish movement pattern. The appearance similarity score, prediction score, and IOU-based similarity scores are combined to produce fish trajectories using the Hungarian algorithm. This framework is capable of reducing the number of ID switches by a margin of 35.6\% on Fish4Knowledge dataset and 3.8\% on GMOT-40 fish category while maintaining high tracking accuracy. This repository is an implementation of the paper titled "Multi-fish tracking with underwater image enhancement by deep network in marine ecosystems" from Signal Processing: Image Communication journal. If you find our work useful in your work, please cite our work: **Prerana Mukherjee, Srimanta Mandal, Koteswar Rao Jerripothula, Vrishabhdhwaj Maharshi, Kashish Katara, “Multi-Fish Tracking with Underwater Image Enhancement by Deep Network in Marine Ecosystems,” in Signal Processing: Image Communication, Vol.138, pp.117321, 2025.** ``` @article{MUKHERJEE2025117321, …