Species-level image classification with convolutional neural network enables insect identification from habitus images.

Oskar L P Hansen,Beulah H Garner,Kent Olsen,Jens‐Christian Svenning,Toke T Høye,Benjamin W Price,Steen Dupont,Alexandros Iosifidis

doi:10.1002/ece3.5921

Abstract

Changes in insect biomass, abundance, and diversity are challenging to track at sufficient spatial, temporal, and taxonomic resolution. Camera traps can capture habitus images of ground‐dwelling insects. However, currently sampling involves manually detecting and identifying specimens. Here, we test whether a convolutional neural network (CNN) can classify habitus images of ground beetles to species level, and estimate how correct classification relates to body size, number of species inside genera, and species identity.We created an image database of 65,841 museum specimens comprising 361 carabid beetle species from the British Isles and fine‐tuned the parameters of a pretrained CNN from a training dataset. By summing up class confidence values within genus, tribe, and subfamily and setting a confidence threshold, we trade‐off between classification accuracy, precision, and recall and taxonomic resolution.The CNN classified 51.9% of 19,164 test images correctly to species level and 74.9% to genus level. Average classification recall on species level was 50.7%. Applying a threshold of 0.5 increased the average classification recall to 74.6% at the expense of taxonomic resolution. Higher top value from the output layer and larger sized species were more often classified correctly, as were images of species in genera with few species.Fine‐tuning enabled us to classify images with a high mean recall for the whole test dataset to species or higher taxonomic levels, however, with high variability. This indicates that some species are more difficult to identify because of properties such as their body size or the number of related species.Together, species‐level image classification of arthropods from museum collections and ecological monitoring can substantially increase the amount of occurrence data that can feasibly be collected. These tools thus provide new opportunities in understanding and predicting ecological responses to environmental change.

Highlights

Recent reports suggest that insect biomass and abundance have been declining dramatically in recent decades (Agrawal & Inamine, 2018; Hallmann et al, 2017; Lister & Garcia, 2018; Loboda, Savage, Buddle, Schmidt, & Høye, 2018; Seibold et al, 2019; Wagner, 2019), even though trends vary if measured across or on individual habitats and species (Loboda et al, 2018)
Setting a minimum acceptable confidence threshold to 0.5 before decreasing taxonomic resolution by one hierarchical level, 75.8% of a total of 19,164 images were classified correctly to the decided taxonomic level and average classification recall across all specimens increased to 74.6%
Within the tested species of British Carabidae, 51.9% of the 19,164 images were classified to the correct species, when testing the model classifying to species level, and 74.9% to the correct genus, using the same trained model with genus names from ground truth and predicted species

Summary

| INTRODUCTION

Recent reports suggest that insect biomass and abundance have been declining dramatically in recent decades (Agrawal & Inamine, 2018; Hallmann et al, 2017; Lister & Garcia, 2018; Loboda, Savage, Buddle, Schmidt, & Høye, 2018; Seibold et al, 2019; Wagner, 2019), even though trends vary if measured across or on individual habitats and species (Loboda et al, 2018). We test the ability of a convolutional neural network (CNN) to classify ground beetles (Coleoptera: Carabidae) to genus, species, or higher taxonomic level from images of specimens within the British collection at the Natural History Museum, London This collection provides a good test case as it has been well curated and assessed for correct species identity, represents a commonly prepared type of insect collection for which this method is directly applicable to, and has access to the SatScan® (SmartDrive Limited; Blagoderov, Kitching, Livermore, Simonsen, & Smith, 2012; Mantle, LaSalle, & Fisher, 2012), a rapid whole drawer imaging system. These prepared specimens can serve a simplified model for what a camera trap would record These images represent a good indicator of the potential taxonomic resolution of automatic species identification with current state of the art classification methods, based on data from a camera trap, when compared to expert identifications of the specimens. To increase accuracy and to critically assess reliability, we postprocess the output and apply thresholds on confidence values for each of the included taxonomic levels to avoid low confidence in predictions

| MATERIALS AND METHODS

Findings

| DISCUSSION