Automated Classification of Animal Vocalization into Estrus and Non-Estrus Condition using AI Techniques

Automatic classification of animal vocalizations, especially for estrus identification, plays a crucial role in wildlife monitoring and conservation efforts. A comprehensive study on the use of machine learning and deep learning models for the classification of animal vocalizations is presented in this paper, with a focus on estrus vocalizations. It explores a dataset of buffalo vocalizations, including distinct patterns associated with estrus. The Mel-frequency Cepstral Coefficients (MFCC) technique is used for feature extraction, which captures the unique frequency content and temporal dynamics of vocalizations. The models employed in this study include Support Vector Machines (SVM), Naive Bayes, Random Forest, K-Nearest Neighbors (KNN), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Convolutional Recurrent Neural Networks (CRNN), and ResNet34. The models are trained on the extracted features and evaluated using appropriate metrics, such as accuracy, precision, recall, and F1. The results demonstrate the effectiveness of machine learning and deep learning models in accurately classifying animal vocalizations for estrus identification. The CNN and RNN models, which directly process the raw MFCC features, show remarkable performance in capturing the intricate patterns of estrus vocalizations. The machine learning models also achieve competitive performance, indicating the importance of these features in distinguishing estrus vocalizations from other classes. Furthermore, automated estrus identification techniques offer numerous benefits, including increased efficiency, reduced human bias, and scalability for large-scale wildlife monitoring projects. The successful application of machine learning and deep learning models demonstrates their suitability for automated estrus identification and opens avenues for further research in this domain. Future directions may include exploring transfer learning approaches, ensemble modeling techniques, and expanding the dataset to encompass a broader range of species and vocalization patterns. The integration of these automated techniques in wildlife monitoring efforts can enhance conservation strategies, enabling timely and informed decision-making for the preservation of endangered species and their ecosystems.