An Automated Vertebrate Animals Classification Using Deep Convolution Neural Networks

Abstract

On over years, the accuracy level of any algorithm for animal detection using a computer vision system is still practically unusable under uncontrolled environment. A lot of interesting has been shown to object detection, recognition, and classification, etc. Visual monitoring in scenes, for animal, is currently one of the most active research topics in computer vision (CV). In spite of there are a lot of research, intelligent, real-time, but the methods of dynamic object detection and recognition are still unavailable. This paper suggests using Deep Convolutional Neural Network (CNN) to detect and classify the animals (vertebrate classes) in digital images. Our dataset consists of 12000 different images, 9600 images for training stage, and the rest images (2400) for evaluation stage. After apply the proposed system, we found the best image size for this algorithm is 50x50 and the best number of epochs is 100. The total performance of the results reached to 97.5%. The experimental results reflected that the proposed algorithm has a positive effect on overall animal classification performance.