TUNADECS: Tuna Detection and Species Classification in Underwater Environment Using Deep Learning

Tuna is a high-value fish species vital to the global commercial fishing industry and Indonesia’s economy. However, declining tuna populations due to overfishing and environmental challenges highlight the need for advanced monitoring and management solutions. This study introduces TUNADEC, a deep learning-based system utilizing the YOLOv8 framework for the detection and species classification of tuna in underwater environments. The system was developed using a custom dataset of 915 annotated bounding boxes across four tuna species: Bigeye Tuna, Bluefin Tuna, Longfin Tuna, and Yellowfin Tuna. Preprocessing steps, including advanced augmentation techniques, were applied to enhance data diversity and robustness. The model’s accuracy, recall, and F1 scores exceeded 90% across all species, indicating its reliability and effectiveness. Performance evaluation on a test set confirmed high accuracy and robust classification capabilities, with minimal misclassifications. Additionally, the model’s confidence scores, particularly for Longfin Tuna at 94.00%, underscore its potential for real-world applications. Despite its strengths, limitations such as dataset size and the lack of integration with IoT systems suggest opportunities for further enhancement. Expanding the dataset and enabling real-time monitoring through IoT integration could amplify its impact on fisheries management and conservation. This study establishes TUNADEC as a valuable tool for sustainable fishing practices, contributing to the digital transformation of fisheries and aiding the preservation of marine ecosystems.