Artificial Intelligence Recognition of Wildlife Species From Camera Trap Imagery

Abstract

Introduction: Camera trapping has now become an indispensable method of non-invasive wildlife monitoring that creates huge amounts of imaging data that require considerable time and resources to process manually. The use of artificial intelligence (AI), specifically, computer vision based on deep learning, becomes a promising option to automate the classification of wildlife species on the camera trapping imagery. Materials & Methods: The current research is based on a large set of camera trap images from different habitat areas that include various wildlife species taken under different environmental conditions such as lighting, weather, and background. Pre-processing of the images involved noise removal, cropping, and augmentation techniques. Several CNN models, including transfer learning-based models, were trained and validated on images to perform the classification task. Results: The recognition models using AI demonstrated high accuracy of classifications in most categories of species. Their efficiency was greatly enhanced with the help of transfer learning and data augmentation techniques. They allowed for making distinctions between visually similar species and performed reliably in adverse situations such as partial occlusion, infrared nighttime imaging, and cluttered surroundings. Comparison has shown that deep learning techniques perform better than traditional image processing and manual identification in terms of both speed and accuracy. Conclusion: The incorporation of AI techniques into the wildlife monitoring system based on camera traps helps to save much effort in species identification and increases its reliability