Beaver Habitat Terrain Identification Using Aerial Imagery

Beavers are important species with a strong influence on ecosystems, their damming activities transform ecosystems through wetland creation and consequent hydrological change. As such they are known as ecosystem engineers. This study concerns the real-world problem of understanding the distribution of beavers by identifying their habitat and terrains using drone imagery. Experiments are performed to investigate several state-of-the-art Convolutional Neural Networks ranging from the custom designed and pre-trained classification models such as ConvNeXt to segmentation models such as Mask RCNN to identify beaver habitats (including dams and lodges) and different terrains. Efforts are made to improve the explainability of the models through spatial heatmaps overlays. This research offers accurate identification of beaver habitats using aerial images by achieving high accuracy rates of 98% which is crucial for effective decision-making related to beavers and understanding the changes that their populations create.