Spider species classification using Vision Transformers and Convolutional Neural Networks

Abstract

Spiders often seek shelter in the heat and safety of homes and although most of them are harmless, some can represent a real danger. Thus, since differentiating spider species can be a challenge for people with no prior knowledge, having a way to identify them could be useful to avoid those that are poisonous. To solve this question, in this project we propose to study and compare convolutional neural networks (CNN) and vision transformers (ViT) regarding the performance in the task of recognizing spider species from their images. These techniques perform automatic feature extraction and learn high-level representations of images, and are currently widely used, for example, in object detection and image classification. Specifically, we aim to evaluate the potential of these networks for the task of classifying spider species, including quantitative and qualitative performance comparisons between the two types of networks. For this, we will use INaturalist's arachnid photo database. The results of this project may contribute to the development of applications aimed at identifying spiders and providing information of interest about the species.