Detection of Wildlife Animals Based on Transfer Learning Using ResNet Algorithm

In order to maintain a balanced eco-system, wildlife detection and monitoring play a crucial role. This paper uses a transfer learning approach to reuse ResNet, which was trained on the ImageNet dataset, to identify and classify wild animals. The ResNet deep learning architecture uses a technique called skip connections to overcome the problem of vanishing gradients. For training purposes, images are collected from kaggle and images.cv dataset, and for testing purposes, images are collected in different locations using the camera. By varying the batch size and epochs, ResNet50V2, Resnet101V2, and Resnet152V2 are compared. ResNet152V2 achieves the highest accuracy of 89.33 for 50 epochs for the own datasets, which have various environmental challenges.