A Deep Multi-View Learning Approach for Wildlife Animal Detection and Identification

Detecting animals and recognizing them is a challenging task in computer vision. The main aim of this paper is to detect the animals in an image from the selected datasets and assign a label to each of the corresponding images. The images are selected from different databases under different conditions. In this work, detection and recognition are done with the help of Principal Component Analysis (PCA) and a Deep residual network, which can overcome error rates in the existing methods due to high false positive and negative rates. The proposed work is compared with existing methodologies such as Convolutional Neural Network (CNN), Support Vector Machine (SVM) and Multi- Layer Perceptron (MLP). In the present approach, PCA is used to create views in the given image for identification and deep residual network is used to alleviate the lack of labeled training samples in order to identify a learned latent space. The results are demonstrated and analyzed on varied datasets using different evaluation metrics. In comparison to other algorithms, the proposed work exhibits enhanced accuracy by 2.27-4.00%.