Treffer: Application of transfer learning in the classification of three bovine breeds using artificial neural networks.

Introduction: The use of Transfer Learning allows leveraging previously acquired knowledge by a model in one task and adapting it to another related task, for example, in the classification of bovine breeds from images. Objective: To develop a precise and efficient bovine breed classification model using Transfer Learning and convolutional neural networks. Methods: 517 images were collected from internet sources for each of the three bovine breeds analyzed: Brahman, Guzerat, and Holstein. These were organized into separate folders for each bovine breed and processed using the TensorFlow ImageDataGenerator library in Python to augment the dataset and improve model generalization; pixel normalization was performed to range between 0 and 1, and 20% of the data was assigned for validation while the rest was used for training. Transfer Learning was employed using the pre-trained MobileNet V2 model. The model was trained with a dense output layer to provide classification results. Once trained, it was incorporated into an interactive web page designed with the Streamlit package, allowing users to upload images and make breed predictions. Results and Discussion: The model showed an increase in accuracy and a decrease in loss function throughout training, reaching 93% accuracy and 0.17 loss. Evaluation of the model with the validation dataset demonstrated 94% accuracy and 0.17 loss. Utilizing a pretrained model as a starting point yielded promising results with a relatively small dataset. Conclusion: The use of Transfer Learning in bovine breed classification through convolutional neural networks is a viable and effective technique that can be implemented in practical applications using images. [ABSTRACT FROM AUTHOR]