Fish Detection in Maritime Surveillance Using Deep Learning
https://doi.org/10.5281/zenodo.19363872
Keywords:
Underwater Object Detection, Fish Detection, Maritime Surveillance, Deep Learning, YOLOv10, RT-DETR, Weighted Box Fusion, Multi-Object Tracking, Computer VisionAbstract
Precise monitoring of the fish population is crucial in conservation of marine ecosystems, sustainable fisheries as well as the maritime surveillance. The underwater detection is a problem because of low visibility, distortion of colors, and complicated backgrounds. In this paper, a Hybrid Deep Learning Framework (HDLF) will be proposed to detect underwater fish, which will be based on the fusion of YOLOv10 in real-time convolutional detection and RT-DETR in transformer-based contextual reasoning. Forecasts are combined through the use of Weighted Box Fusion (WBF) to enhance the accuracy of the bounding box. Compared to single models, the hybrid framework is superior in terms of precision, recall, and mAP values, and works in real-time on large public underwater datasets.
