A Web-based Application for Snake Species Identification using Vision Transformer and CNN-based Ensemble Meta Classifier

Abstract

Being a tropical country, Sri Lanka has one of the highest snakebite rates in the world. In 2019, 50 snake bite fatalities have been reported in Sri Lanka. Therefore, the accurate identification of the snake category is crucial for healthcare workers to diagnose and treat the victims as well as to save the snake from being killed. In this paper, we present a web-based application based on Convolutional Neural Network and Vision Transformer architectures to classify between the Russell's viper and the Indian Rock Python. Five different image classification models were trained using the pre-trained architectures ResNet-50, ResNet-100, EfficientNet B0, EfficientNet B7 and Data-Efficient Image Transformers. We were able to gain a testing accuracy of 94.5% by using an ensemble approach for the mentioned classifiers. Furthermore, this study presents the first web-based application in Sri Lanka enabling the automatic identification between Russell's Viper and the Indian Rock Python.