Hierarchical classification of pollinating flying insects under changing environments

Abstract

Automatic monitoring of flying insects enables quick and efficient observations and management of ecologically and economically important targets such as pollinators, disease vectors, and agricultural pests. Studies on this topic mainly cover the tasks of detection and identification or classification, the latter often guided by the flight sounds of insects. This paper uses domain knowledge and taxonomy information to classify bee and wasp species based on abiotic variables and wing-beat data that change depending on climatic-environmental conditions. We survey the state-of-the-art in hierarchical classification and evaluate the most popular local and global methods for this task on flight data from nine hymenopteran species. We collected the data in Brazilian fields employing an inexpensive optical sensor. Our results show that it is possible to hierarchically classify groups of specimens per species, species, and groups of species according to their wing-beat data at different temperature and relative humidity levels with at least 91% accuracy. Besides benefiting research aimed at building insect classifiers adaptable to natural variations in the environment, this study is a vital step in a series of efforts to design non-invasive species monitoring techniques.