INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 8, ISSUE 10, OCTOBER 2019 ISSN 2277-8616 2613 IJSTR©2019 www.ijstr.org Recognition Of Animal Species On Camera Trap Images Using Machine Learning And Deep Learning Models * Rajasekaran Thangarasu ,Vishnu kumar Kaliappan ,Raguvaran Surendran,Kandasamy Sellamuthu ,Jayasheelan Palanisamy Abstract: Wild animal movement monitoring and its distribution are essential for the conservation of animal life. Camera trap, a most commonly used technique for animal monitoring which automatically activate the camera on animal presence and obtain a huge volume of data. The present work aims to investigate various machine learning algorithms including Support Vector Machine (SVM), Random Forest (RF) and deep learning models such as Alexnet, Inception V3 for classification of animal species. Among which deep learning models outperforms than machine learning algorithms. In this paper, the overall comparison of accuracy between machine learning and deep learning models has been observed and discussed. The outcomes of the experiment suggest that InceptionV3 attains more accuracy than SVM, Random Forest, AlexNet and also results highly accurate classification is obtained with the availability of enough data and precise techniques. The experiment uses KTH dataset that composed of 19 different categories of animals among which 12 classes are selected to measure the performance of the models. Keywords: Animal species recognition, camera trap, KTH dataset, Random Forest, SVM, AlexNet, Inception v3, fine-tuning ———————————————————— 1. INTRODUCTION Wildlife monitoring is crucial for tracking animal habitat utilization, population demographics, poaching incidents, and movement patterns. Numerous technology has been introduced that includes motion-sensitive camera traps, radio tracking, wireless sensor network tracking, and satellite for monitoring wild animals 1-2 . Currently, the animal espial and recognition are still an arduous challenge and there is no unique method that provides a sturdy and efficient solution at all situations. Monitoring wild animals through camera traps are prominent due to their commercial availability, equipped features, and ease deployment. The Extraction of knowledge from these camera-trap images is implemented using machine learning and deep learning models. Machine learning (ML) plays a key role in a wide range of statistical, image recognition, natural language processing, and expert systems. Every instance in the dataset is represented using the set of features and class labels where the training data builds a predictive model. Some of the ML algorithms for feature extraction and classification are Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), and Random Forest. In this paper, Random forest and SVM has been used for the classification of wild animals using camera trap images. A proposed CNN model, Deep learning (DL) outperforms in the identification of wild animals using camera trap images without any manual intervention 15 . Most popular deep learning architectures are GoogleNet, AlexNet, VGG, ResNet, NiN, and inception-1, 2, 3 that are employed in the classification of images 3-4 . Among these architectures, Alexnet and Inception V3 are used for the classification of animal species images in KTH dataset. The level of accuracy at each method has been calculated and compared. The outline of this paper is organized as follows. Section 2 gives a brief overview of the related works in animal identification. In Section 3 the details about the datasets used for experiment purpose followed by In Section 4 the animal identification based on feature extraction and classification is discussed. Section 5 the obtained experimental results are listed. Finally, Section 6 Concludes and suggests future work. 2. RELATED WORK Gullal Singh Cheema and Saket Anand developed an automatic wildlife monitoring system for detection and recognition of individual animal species like tigers, zebras, and Jaguars which are in different pattern. The faster Regional CNN is proposed to detect the whole body and the flank region of individual animals. The features and flanks extracted from animals are fed to the AlexNet to train the logistic regression for identification of individual 16 . Olivier Chapelle, Patrick Haffner, and Vladimir N. Vapnik developed Support Vector Machines for Histogram based Image Classification. It describes that the support vector machines (SVM’s) are used to generalize well on difficult and different image classification problems where the only features which are used are high dimensional histograms 17 . Tibor TRNOVSZKY, Patrik KAMENCAY and Richard ORJESEK implemented the Animal Recognition System Based on Convolutional Neural Network. The main goal of this paper was to compare the overall recognition accuracy of the PCA, LDA, LBPH, and SVM with proposed CNN model. Mohammad Sadegh Norouzza deh demonstrated the state-of-the-art deep learning neural network method which results in automatic identification of animals on Snapshot Serengeti dataset which had of 3.2 million images with 42 different species of animals. Various deep learning architectures such as AlexNet, NiN, VGG, ResNet-18, 34, 152 were used to classify animal species. Among the six architectures, VGG provided a better accuracy of 96.8 % 7 . ————————————————  T. Rajasekaran, K.Vishnu kumar S.Raguvaran ,S.Kandasamy are Cyber Physical System Group and Professor in Department of Computer science Engineering,KPR Institute of Engineering and Technology, Coimbatore,Tami nadu,641407,India. E-mail: phdresearchpaper@outlook.com  P.Jayasheelan, is Professor in Department of Computer science,Sri Krsihna AdthiyaCollege of Arts and Science,Coimbatore, Tamilnadu.