Transfer learning application on snake classification

Abstract

Snake bite envenoming is a public health issue that is often neglected in many countries, claiming approximately 100,000 lives annually. Currently, there is a mobile application available on the Android platform that can be used to classify snake from a given image. However, it requires manual identification and classification by snake experts, which may take an indeterminate amount of time. Therefore, there is a need to develop a mobile application that is able to classify snake images quickly and accurately. This is possible by applying machine learning methods to classify the snake images and creating a simple mobile application for public use.

By applying transfer learning method on some pre-trained models, the author have successfully achieved 89\% validation categorical accuracy in snake species classification. This level of accuracy is comparable to that of human observers, and as such it is reasonable to say that this model is now ready for real-life usage. A simple mobile application is then developed to access the model for classifying snake images, thus possibly saving many lives in the process. This application has also enabled the crowd-sourcing framework to gather more snake images, which can be used to improve the model in the future.

The author has also conducted an online survey based on technology acceptance model (TAM) to study the factors which influence the users to use our application. Overall, the study finds that the users perceived our application as useful and easy to use. They have also shown a positive behavioral intention towards using the application. There are also other interesting findings that should be investigated further to gain more insight on this area of research.