Image-Based Identification of Animal Breeds Using Deep Learning

Abstract

Accurate and reliable breed identification of domestic animals from images is one of the most promising but challenging tasks in intelligent livestock management. Traditional methods for animal breed identification are very costly and time consuming. Therefore, there is a need for a faster and cheaper technique for animal breed identification, which can be used by anyone without much technical knowhow. Deep Learning based animal breed classification from images can be used to solve this problem. Recent developments in deep Convolutional Neural Network (CNN) has drastically improved the accuracy of image recognition systems, but choosing the optimal model for the required task is very important for best performance. In this study, the performance of nine different deep CNN-based models have been analyzed to find the optimal model which can precisely determine the breed identity of individual animals from its image. All nine CNN models have been separately trained end-to-end on Pig Breed Dataset and Goat Breed Dataset using a set of identical hyperparameters. From the results obtained it has been established that MobileNetV2 is the best deep-CNN model for Goat Breed Classification with 95.00% prediction accuracy and InceptionV3 is the best model for pig breed classification with 100.00% prediction accuracy. Breed classification performance of goat and pig obtained in this study have been compared with other techniques used for animal breed classification. Comparison results show that our CNN-based technique has performed on par with all other methods. With these encouraging results, it can be confidently stated that deep CNN-based models can be used for solving the animal breed classification problem with high accuracy and can be used as ready to use technology for intelligent livestock management.