Machine learning for image based species identification

Waldchen and Mader reviewed the state of automated species identification from images across taxa, covering the progression from hand-crafted features to deep learning and the datasets, benchmarks, and accuracy levels achieved. While broader than camera trapping, the review situates camera-trap classification within the wider field of biological image recognition and distills lessons about what makes identification easy or hard, including intraclass variation, similar species, and image quality. It surveys applications from plants to animals and discusses the practical requirements for building reliable identification systems. As a comprehensive overview of image-based species recognition, it provides useful context for camera-trap researchers adopting machine learning. The paper is widely cited as a reference on the capabilities and limitations of automated species identification.