Snow Leopard Recognition Using Deep Convolution Neural Network

The paper describes the use of Deep Convolution Neural Networks (DCNN) for the recognition of Snow Leopards, from a data set of photos taken in the wild. The data set comprises of 1500 images, captured in the Himalayas using motion sensing cameras. The images contain numerous living species, ranging from a butterfly to a human being, other than Snow Leopard. For the training phase we divided the data set into two classes, Snow Leopard and Other Animals. The Snow Leopard class contains photos showing more than one animal, from different angles, having different sizes, body parts because of distance from camera and several backgrounds. The photos are converted to 200 x 200, grey scale images in the preprocessing phase. A 5 layer DCCN, constituted of 3 convoluted and 2 fully connected layers, is employed for the experimental setup. Rectified Liner Units (ReLU) is used as the activation function in the fully connected layers and softmax function is applied for classification. The evaluation of the system shows an overall 91% accuracy, along with sensitivity of 0.90 and specificity of 0.88 for Snow Leopard class identification.