Application of artificial intelligence technologies in processing images from camera traps: principles, software, approaches

Artificial intelligence (AI) is increasingly penetrating environmental science. This is most rapidly evident in image-based research, for example from drones or camera traps. This review discusses the current development of camera trap research in the application of AI technologies, namely computer vision and deep learning. The basic concepts of Machine Learning (ML) and deep neural networks are briefly considered, which are essential for the modern biologist and ecologist to understand the images processing and analysis. The possibilities of using AI for patterns recognition and object search in images are discussed. An overview of modern software using computer vision and machine learning technologies for recognizing photos and video from camera traps is provided, as well as a brief overview of open data sets for training ML models. In general, eight ML-software are considered: MegaDetector, EcoAssist, MLWIC2, Conservation AI, FasterRCNN+InceptionResNetV2, DeepFaune, ClassifyMe, as well as the first domestic development from the Moscow Institute of Physics and Technology. On the basis of research with camera traps in the Central Forest Nature Reserve, the software was tested and its advantages and disadvantages were identified. In conclusion, the potential of AI application in modern research with camera traps is discussed. This review will be useful for both biologists and ecologists for general acquaintance with neural networks and their application in the field of pattern recognition on images, and for ML-specialists and programmers to understand the applicability of ML models in environmental science and the possibilities of their training on large data sets.