Image-Based Individual Fish Identification as a Substitute for Invasive Methods: State-of-the-Art and Research Gaps

Individual identification of fish within the same species is one of the most crucial and frequently performed tasks in aquaculture research, and it is also widely used in aquaculture production. Fish sampling, treating, welfare, and disease state estimation are increasingly turning from group-based to individualized. Traditionally, fish individual identification has relied on marking methods such as PIT tagging, which are invasive, time-consuming, and labor-intensive. It is also very constrained in large cultivation units such as cages. Consequently, there is a demand for developing non-invasive techniques that are rapid and cost-effective. Image-based individual identification has the potential for developing remote, faster, and more economical methods for in situ identification of individual fish. This article reviews the development of these image-based methods for fish individual identification over the past decades, presenting their fundamental concepts and principles. It analyses the strengths and weaknesses of each approach and outlines future research directions. Recent studies indicate that the application of convolutional neural networks holds great promise in accelerating the development of new techniques for more effective fish identification. However, the accuracy, large amount of training data, and specific experimental conditions of these methods still need improvement to meet the demands of intensive aquaculture. Through close collaboration between fisheries experts and engineers, the precision and level of intelligence in fish individual identification techniques will be further enhanced based on the aforementioned methods.