Deep Learning Based Classification of the Big Four Snake Species Using Visual Features

According to a recent study conducted by the Centre for Global Health Research and published by the World Health Organization (WHO)., between 81.,000 and 138,000 deaths occur from snakebite envenomation annually. Approximately 45,900 snakebite deaths occur in India alone each year; of these, ninety percent are due to the “Big Four” snake species, which are the Indian cobra (Naja naja), the Sawscaled viper (Echis carinatus), the Common krait (Bungarus caeruleus), and the Russel's viper (Daboia russelii). The majority of these cases occur in rural areas, where identifying the snake responsible for the bite quickly is critical for saving victims' lives. Automation for identifying the species of snakes has significant potential to enable timely and customized medical care, ultimately decreasing the number of fatalities. This research focuses on the exploration and evaluation of various computer vision algorithms such as Xception, EfficientNet, DenseNetl21, MobileNet V2, Inception V3, Vision Transformer (ViT), ResNetV2 to successfully classify the ‘big four’ snake species. The research then presents a performance comparison based on the experimental findings, employing a K-Fold Cross-Validation approach to gather performance results for each model. The results indicate that Vision Transformer (ViT) outperformed other models, achieving 91.1% accuracy rate with 0.9 as the macro-F1 score. The results highlight the efficacy of deep learning algorithms for snake species identification, providing a promising avenue for accelerated treatment tailored to the specific snake species involved.