Evaluating Image-based Species Recognition Models suitable for Citizen Science Application to Support European Invasive Alien Species Policy

Recent developments in image recognition technology and its application to automated species identification led to an increase in the research of computer vision models. These models play a growing role, especially for the detection and tracking of Invasive Alien Species (IAS) as one of the main drivers of biodiversity loss globally. Here, Citizen Science (CS) is a very promising and already successful approach of involving the public in IAS recording with the help of mobile applications (apps). However, these apps often use computer vision models specialized for distinct classes of organisms or habitats, but not for locally relevant invaders. Our work evaluates image-based species recognition models suitable for use in CS apps to meet the purposes of the European Invasive Alien Species policy. The report includes a state of the art analysis of current species recognition models. It describes a methodology for testing selected models against the IAS list of union concern, a candidate list, and local lists for European regions. The results show that no existing model could detect all species on the above mentioned lists, but several models, such as the iNaturalist API and the Microsoft AI for Earth model, show high accuracies throughout different classes of organisms. The report closes with recommendations on the future use of these models in CS apps - by either collaborating with model providers to add missing species, or by training open source models with additional image data to meet the European purpose.