Gender determination of domestic chicks based on vocalization signals

Gender determination in chicks is an important task in poultry production and is helpful for precision feeding of different sexes. At present, most chicken sex identification methods need to be completed manually by pro-fessionals, which is time-consuming and laborious. In this paper, a method was designed to determine the sex of one-day-old chicks according to vocalizations. This method uses sound technology to detect chick vocalization and automatically detects vocalization endpoints by the double threshold method using three parameters: short-term energy, short-term zero crossing rate and duration. The audio features were extracted as the input of the three deep learning models for learning and classification. In the experiment, the training set was used to train the model, and the test set was used to calculate the detection results. The vocalizations of the training set and test set came from different chicks. In the gender detection of each vocalization, the accuracy of convolutional neural networks (CNN), long short-term memory (LSTM), and gate recurrent unit (GRU) was 74.55%, 75.73% and 76.15%, respectively. The highest recall was 77.03% of GRU, and the highest specificity was 78.38% of LSTM. After that, the gender of chicks was predicted according to the vocalization detection results. The average accuracy values of CNN, LSTM and GRU for vocalization were 91.25%, 87.08% and 88.33%, respectively. The experimental results show that the method proposed in this paper can be used to detect the gender of chicks by vocalization, which is of great significance for automatic chick gender detection and intelligent poultry production.