A multi-scale temporal feature fusion framework for sheep voiceprint recognition

Abstract

Voiceprint recognition technology is an effective way to identify individual sheep; however, related research is lacking. To this end, we propose a hybrid model based on the ResNet18 network and gated recurrent units (GRUs) to comprehensively represent the input data. The model uses the feature pyramid network (FPN) structure and a one-dimensional convolutional block attention module (1D-CBAM) for feature fusion to enhance the classification ability of the model. This model is used to extract sheep voiceprint features and combined with the proposed similarity correction method to construct a sheep voiceprint recognition system. The model is trained on a dataset including 300 sheep from three different breeds. The results of 5-fold cross-validation experiments show that the average recognition accuracy (Acc) and average contrast accuracy (CA) of the model reach 98.86 % and 98.66 %, respectively, with an average equal error rate (EER) of 1.34 %, demonstrating that the improved method is stable and reliable for sheep voiceprint recognition. This study provides a new solution for the identification of individual sheep.