Review on Automatically Identification of Multi Wild Animal Species in Camera Trap images using Deep Convolution Neural Network

Abstract

Technology plays an important part in wildlife and ecosystem conservation, and can vastly reduce the time and effort spent in the associated tasks. Having accurate, detailed, and up-to-date information about wildlife species is the challenging task. Presently such data are gathered manually with great expense. In recent years, motion sensor cameras called camera traps enable pictures of wildlife to be collected inexpensively at high volume. Such cameras are remote, independent devices, triggered by motion and infrared sensors that provide the images of passing animals Automated camera-traps used in wildlife studies are small boxes, secured to a tree, rock or other structure. Camera-traps are used to detect rare species, delineating species distributions, and monitoring animal behavior. Although camera trapping is a useful method in ecology, this method generates a large volume of images. Therefore, it is a big challenge to process the recorded images and even harder, if the biologists are looking to identify all photographed animals instead of looking for a certain species. Currently, no automatic approach is used to identify species from camera-trap images. Researchers and citizen science volunteers analyze thousands or millions of photographs manually. Automatic classification of animal species in camera-trap images still remains an unsolved problem due to very challenging image conditions To address these issues, in this work automatic classication of camera trap images has been a focused. Recent advances in artificial intelligence have enabled researchers to improve automatic species identication signicantly. To support wild animal detection researchers have developed various automatic and semiautomatic algorithms. Among these algorithms, deep learning techniques achieve outstanding performances in wild animal detection. In particular, pre-trained CNN models have proven to be particularly successful at classifying and detecting objects in images compared to traditional descriptor based strategies Furthermore, by using transfer-learning, our work investigates how to leverage model strained on large camera trap datasets to smaller datasets.