Automatic recognition methods of fish feeding behavior in aquaculture: a review

Abstract

Feeding is a major factor that determines the production costs and water quality of aquaculture. Analysis of fish feeding behavior forms an important part of the feeding optimization. Fish feeding has generally been performed with automatic feeding machines which can lead to excessive or insufficient feeding. Recognition of fish feeding behavior can provide valuable input for optimizing feeding quantity. Due to the complexity of the environment and the uncertainty of fish behavior, the correlation and accuracy of behavior recognition are generally low. The accurate identification of fish feeding behavior till faces substantial challenges. This paper reviews the technical methods that have been used to identify fish feeding behavior in aquaculture over the past 30 years. The advantages and disadvantages of each method under different experimental conditions and applications are analyzed. Many methods are effective at evaluating and quantifying fish feeding intensity, but the recognition accuracy still needs further improvement. It is proposed by this paper that technologies such as data fusion and deep learning has great potential for improving the recognition of fish feeding behavior.