Fish school feeding behavior quantification using acoustic signal and improved Swin Transformer

Abstract

In aquaculture, the real-time quantification of fish feeding behavior is an important basis for feeding decisions. The acoustics produced by fish chewing feed and activities during feeding can be used to quantify feeding behavior. This study proposes an Audio Spectrum Swin Transformer (ASST) model based on an acoustic signal and attention mechanism that can divide the feeding intensity of fish into four grades: strong, medium, weak, and none. The specific implementation methods are: (1) A sliding window was applied to clip the audio and the acoustic. Signals are transformed into spectrograms. (2) The perceptual domain of fish feeding acoustic spectrum recognition task was gradually expanded by adopting the Swin Transformer and utilizing its hierarchical structure. (3) The model's performance for small data sets was further improved by adding SPT, LSA, and enhanced residual connections. (4) A predictive optimization module was designed to correct the feeding strategy according to four feeding levels. The final experimental results demonstrate that the accuracy of the improved ASST network for fish feeding behavior quantification reached 96.16%. It can effectively identify-four fish feeding intensities, allowing on-demand feeding and providing a basis for developing intelligent feeding machines.