Deep Learning-Based Fish Classification with Data Augmentation

Accurate identification and recording of different fish species' distribution and abundance provide deeper insights into local marine ecosystems and aid in the formulation and implementation of effective conservation strategies. However, due to the influence of fish similarity and underwater optical characteristics, the identification of fish needs professional knowledge and experience. Therefore, it is necessary to develop an effective underwater fish species recognition system using artificial intelligence. In this study, we employed the YOLOv8 neural network architecture for model training and recognition. Additionally, various data augmentation techniques such as color channel adjustment, perspective transformation, image overlay, and flipping were used to improve the recognition accuracy for fish species with limited training data. The proposed model achieves a mean Average Precision (mAP) of 93.8% for 30 different fish species.