Computer Vision Models in Intelligent Aquaculture with Emphasis on Fish Detection and Behavior Analysis: A Review

Intelligence technologies play an important role in increasing product quality and production efficiency in digital aquaculture. Automatic fish detection will contribute to achieving intelligent production and scientific management in precision farming. Due to the availability and ubiquity of modern information technology, such as the internet of things, big data, and camera devices, computer vision techniques, as an essential branch of artificial intelligence, have emerged as a powerful tool for achieving automatic fish detection. At present, it has been widely used in fish species identification, counting, and behavior analysis. Nevertheless, computer vision modeling used for fish detection is riddled with many challenges, such as varies in illumination, low contrast, high noise, fish deformation, frequent occlusion, and dynamic background. Hence, this paper provides a comprehensive review of the computer vision model for fish detection under unique application scenarios. Firstly, the image acquisition system based on 2D and 3D is discussed. Further, many fish detection techniques are categorized as appearance-based, motion-based, and deep learning. In addition, applications of fish detection and public open-source datasets are also presented in the literature. Finally, the prominent findings and the directions of future research are addressed toward the advancement in the aquaculture field throughout the discussion and conclusion section.

Automated summary

Intelligence technologies are crucial in improving product quality and production efficiency in digital aquaculture, with computer vision techniques emerging as a powerful tool for automatic fish detection. However, challenges such as variations in illumination, low contrast, and fish deformation are present in computer vision modeling for fish detection.