Research on Underwater Organism Detection Technology Based on YOLO v5

Le résumé fourni par la source

In recent years, deep learning and related technologies have experienced rapid development due to their characteristics of high speed, accuracy, and recognition rates, as well as environmental sustainability. They are increasingly being applied in real-time inspection research involving underwater robots for the automatic identification and classification of underwater targets. This paper builds a self-constructed underwater image dataset, annotates the location information of four types of underwater targets: tortoises, fish, person, and corals, and performs preprocessing such as sharpening, histogram equalization, and normalization on the targets in the sample set to improve image quality. Based on this, a YOLO v5 network model is established. After repeated training, the loss function curve proves that the model is effective, and the average accuracy of the four types of targets is: 97.1% for tortoises, 93.5% for fish, 88.9% for person, and 74.7% for corals. Finally, to verify the reliability of the network model, this paper redivides the training set and test set in the ratio of 7:3 and 8:2 to achieve repeated training of the YOLO v5 network, and evaluates the model performance through three indicators: accuracy, recall rate, and mean average precision. Ultimately, the feasibility and stability of the YOLO v5 algorithm for underwater target detection are verified.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.