Efficient Marine Fish Classification Using Hybrid Deep Learning Model

Illegal, unreported and unregulated fishing practices threaten marine ecosystems, global seafood supplies, and local livelihoods. Unscientific small-scale fisheries in biodiversity hotspots have a significant impact if it is not constrained to catch the size of fish. The computer vision technique has shown that it can process real-time images to monitor and assist humans in various applications. In this study, an experiment conducted on the Nature Conservancy Fish Dataset (NCFM dataset), contains images of 7 fish species. Hybrid deep learning based on YOLOv5 and ResNet152 was used for the efficient classification of different fish species. The proposed system achieved a Mean Average Precision (mAP) of 0.93 on YOLOv5 for detection and an F1 score of 0.98 on ResNet152 for the classification of fish species.