Exploring the potential of deep learning for classifying camera trap data: A case study from Nepal - working paper

Data from camera trap networks provide crucial information on various important aspects of wildlife presence, movement, and behaviour. However, manual processing of large volumes of images captured is time and resource intensive. This study explores three different approaches of deep learning methods to detect and classify images of key animal species collected from the ICIMOD Knowledge Park at Godavari, Nepal. It shows that transfer learning with ImageNet pretrained models (A1) can be used to detect animal species with minimal model training and testing.  These methods when scaled up offer tremendous scope for quicker and informed conflict management actions, including automated response, which can help minimise human wildlife conflict management costs across countries in the region.