Efficient dorsal fin-based classification of Risso's and common Bottlenose dolphins using YOLOv7 and YOLOv8 models for real-time applications

• Published in: International Journal of Advanced Technology and Engineering Exploration Vol. 11; no. 115; p. 875
• Main Authors: Alhashmi, Fawaghy, Alhefeiti, Maryam, Shaher Bano Mirza, Fouad Lamghari Ridouane
• Format: Journal Article
• Language: English
• Published: Bhopal Accent Social and Welfare Society 01.06.2024
• Subjects: Accuracy; Artificial intelligence; Biodiversity; Classification; Computer vision; Dolphins; Dolphins & porpoises; Ecosystems; Feature maps; Habitats; Identification; Marine environment; Marine mammals; R&D; Real time; Research & development; Research centers
• ISSN: 2394-5443; 2394-7454
• DOI: 10.19101/IJATEE.2023.10102512

The existence of whales and dolphins serves as a key sign of the well-being of the marine environment of that area. It is imperative to undertake research and conservation initiatives to safeguard these marine mammals and their ecosystem. This guarantees their persistence for the well-being of future generations. In recent years, marine surveys conducted in Fujairah offshore waters have generated valuable data concerning the distribution of cetacean species. Notably, common Bottlenose dolphins (Tursiops truncatus) and Risso's dolphins (Grampus griseus) have emerged as prevalent species in the region. These data hold significant information that is useful in species identification and its habitat loss mitigation efforts. Computer vision offers an efficient solution for analysing and interpreting vast visual data compared to of the manual detection methods. Therefore, the primary objective of this study is to assess and contrast the efficacy of you only look once version 7 (YOLOv7) and you only look once version 8 (YOLOv8) models in the identification of cetacean species. The findings indicate that both models exhibit strong performance in identifying and categorizing the desired species. Specifically, YOLOv8 demonstrates a slightly superior precision rate of 91.6% compared to YOLOv7. Additionally, YOLOv8 exhibits improved recall (92.5%) and mean average precision (mAP) of 95.9%. The improved performance of YOLOv8 can be attributed to its comprehensive feature map and optimised convolutional network, combined with a novel backbone network.