ABSTRACT
Demand for data about insects is at an all-time high, while new technology can significantly accelerate data collection. Image-based insect monitoring, whether in the field or the laboratory, is particularly promising. Cameras can record insects consistently through time, while automated image processing enables rapid quantification of insect numbers, identities, traits, behaviours, and interactions. Resulting data can improve understanding of status and trends in insect biodiversity, but also the underlying drivers. We introduce three primary approaches to image acquisition: (i) Manual in-situ imaging by volunteers or professionals; (ii) automatic in-situ imaging by autonomous field cameras; or (iii) ex-situ imaging under highly controlled conditions. We then describe methods to automatically extract biological information from images, including computer vision and machine learning for detection, classification, and characterisation of insects. Targeted manual annotations offer valuable insights and are crucial to train, validate, and test models for image processing. Automated visual systems face many challenges, which may be overcome by implementing standards for protocols, data and metadata, and training models that leverage taxonomic hierarchies and background information. Automated visual systems must engage taxonomists, both amateur and professional, and direct their expertise towards tasks of appropriate difficulty. According to these principles, automated visual systems can facilitate a step-change in insect monitoring and evidence-based conservation.
