Snake Species Identification and Safety Advisory System for Residential Areas in Sri Lanka Using Deep Learning Techniques

Abstract

"Sri Lanka's rich snake biodiversity, with 105 species. The majority, 64 species, are non venomous, with the remaining 22 species are highly venomous, 5 as moderately poisonous, and 12 as mildly venomous. Misidentification and fear contribute to the tragic deaths of innocent snakes, disturbing ecological balance and continuing human-snake conflict. The major goal of this research is to find a solution to the complex problem of accurately identifying snake species in residential areas in Sri Lanka, which poses a significant challenge due to its complexity and extreme difficulties. The current identification system in Sri Lanka relies on human effort and invalid methods, often leading to misidentification. This study gathers data from residential areas, considers differences in snake features, and develops an integrated safety advising system. The system will provide first-aid guidelines and emergency service locations. Using Convolutional Neural Networks and Image Processing, Serpify will assist residents in identifying 16 snake species by uploading or capturing images. As for the initial test results, the model is tested on a sample set of images, and the accuracy, loss, precision, recall, and F1-score are provided. The implemented model achieves around 75% accuracy. Learning curves and the confusion matrix are also shown to examine the model's performance and identify error sources. "