Design and Analysis of Artificial Intelligence Model for the Global Issue of Poisonous Reptile Identification

Abstract

It is estimated that between 80,000 and 140,000 people die year from snakebite envenoming, and another 450,000 become disabled due to this neglected tropical disease. By the year 2030, the World Health Organization hopes to have reduced this load by half. To get there, we need to improve healthcare practitioners' access to up-to-date information and diagnostic tools, as well as fill up the data gap in snake ecology and snakebite epidemiology. First, we need better taxonomic identification of snakes that can cause bites. With the availability of AI-based identification tools for other animals, there is a unique chance to apply machine learning to snake identification and the potentially fatal condition of snakebite envenoming. We trained a deep learning model using 386,006 pictures of snakes, including 198 poisonous and 574 non-venomous species across 188 nations, using the cutting-edge neural network architecture Vision Transformer. We used Flickr and the online biodiversity databases iNaturalist and HerpMapper to compile these images.

For the first time, we demonstrate that AI is capable of correctly categorizing a wide range of snakes, both venomous and nonvenomous, from throughout the world, including similar-looking species from snakebite-prone regions. For snakebite epidemiologists and healthcare practitioners, herpetologists, and the general public, this study lays the groundwork for building global, regional, or national snake identification support systems.