Visual search using artificial intelligence (image recognition of fauna species in Singapore : back-end development of mobile application)

• Main Author: Ong, Jim Liang An
• Other Authors: Yap Kim Hui
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/150024
• Institution: Nanyang Technological University

This project recognizes that educational efforts to create a more ecologically conscious society and technological advancement can come together in establishing an eco-friendly Singapore. To protect the remaining biodiversity and promote eco-consciousness in Singapore, this project aims to identify and classify local fauna species by performing fine-grained classification on a cross-platform mobile application. To achieve the goal, this project is carried out in three stages. In the first stage, a hundred thousand images of fauna species are crawled using Flickr API and sorted into 7 distinct superspecies categories, which are amphibian, mammal, reptile, dragonfly, butterfly, freshwater fish, and bird. Subsequently, to solve the fine-grained classification problem, the Attentive Pairwise Interaction Network (API-Net) is utilized to train the classifiers. Thereafter, the trained classifiers will be deployed on AWS Lambda cloud to acquire high graphical computing power and scalability. For the second stage, MERN (MongoDB, Express.js, React Native and Node.js), a full-stack solution will be used to develop a cross-platform mobile application to make use of the trained classifiers. For the front-end development, the User Interface (UI) and User Experience (UX) is designed and build using the React Native framework. For the back-end development, the REST APIs is designed and developed using Node.js and Express.js. Utilizing MongoDB, a document-oriented database is created to store the information of users and fauna species. For the third stage, the front-end and back-end are integrated to recognize local fauna species. The mobile application achieved a high performance, with the best accuracy of 95.25% among the 7 superspecies categories and perform excellently with other useful features. Lastly, a review is done to further improve the performance of the classifiers and expand the features of the mobile application.