Wild Animal Detection using Deep learning
K L Sreedevi, Anitha Edison
Abstract
The collision of vehicles with animals is an emerging threat to humans and wildlife. Efficient observation of wild animals is crucial. Cost-effective techniques for observing the behaviour of wild animals are needed for both wildlife conservation and for reducing human-wildlife conflicts. Therefore an efficient system is required for wild animal detection. Since there are many different creatures, it is challenging task to manually identify each one. Animal detection can help to prevent animal-vehicle accidents and can trace animals. This will be achieved by applying effective deep learning algorithms. The objective is to create an algorithm for detecting wild animals. The depth-wise separable convolution layer, which combines pointwise and depth-wise convolution, is used to build the model. The suggested model adds zero padding in order to maintain edge characteristics and regulate the size of the output image. IWildCam data is used for testing the proposed algorithm. We achieved promising results with intersection over union value as 0.878 for detection and 99.6% accuracy for classification of wild animal.