Abstract
Identification of arthropods is important in academic and medical applications such as species-species interaction studies and identification for medical diagnosis. Deep learning is a tool that can be used to solve these problems quickly and accurately. For this study, a deep learning model was developed that has the capability of identifying North American arthropods to the genus level and compared multiple methods to increase the performance of this model. These methods include changing the neural network architecture, class balancing, and changing the image input size. The full deep learning model using InceptionResNetV2 obtained top 1 accuracies of 80% and top 5 accuracies of 92%. Comparatively, it was found that changing the neural network to EfficientNetB7 in a subset of the full model achieved a top 1 accuracy of 90%. It was also found class balancing in certain circumstances increased recall and that increasing image input size had a logarithmic effect on performance.