ABSTRACT
This paper presents an approach for Philippine bird species classification using Deep Learning specifically Transfer Learning. We have developed a Philippine bird species dataset since most existing bird species dataset are limited to birds from certain regions such as Northern America. The performance of Convolutional Neural Network (CNN) on image classification has been shown to be very efficient but requires a very large dataset and a significant amount of training time in order to generate a very high accuracy in its classification. In Transfer Learning, the pre-trained model output of the CNN can be further fine-tuned in order to classify image datasets that could possibly be different from the original dataset. In this paper, the pre-trained model of Google’s Inception-v3 having 1000 different categories is retrained to classify 50 Philippine bird species using the TensorFlow library. Our approach achieved an accuracy of 94% for the Philippine bird species classification.
