Visual analytics Usiug artificial intelligence (image recognition for fauna species in Singapore : front-end development)

• Main Author: Goh, Jin Wei
• 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/10356/149028
• Institution: Nanyang Technological University

This project is a full-stack mobile application development project, and it is divided into three stages, which are deep-learning model development, back-end development, and front-end development. The main objective of the project is to identify and classify fauna species in Singapore by performing fine-grained image classification on a mobile phone. In the first stage of the project, a hundred thousand fauna species images are crawled from Flickr and they are classified into seven categories, which are birds, butterfly, dragonfly, mammal, reptile, amphibian, and freshwater fish. To differentiate fauna species family precisely, fine-grained image classification with Attentive Pairwise Interaction Network (API- Net) framework is implemented to train the model by using PyTorch library. For the next stage of the project, a cross-platform mobile application is proposed and developed to fully utilise the functionality of the trained classification model. The classification model will be deployed on AWS Lambda cloud to have high graphical computing power and further reduce the workload of the mobile phone. For user data management, MongoDB is used to create a document-oriented database and enable users to have sign-in authentication and save the data information of fauna species. For front-end development, the user interface of the cross-platform mobile application is designed and developed by using React Native framework. Therefore, the mobile application can run smoothly on both iOS and Android platforms by using the same code structure. Before the development stage, a mobile application workflow is planned to have simplicity and consistency in the workflow design to help users achieve their goals efficiently and further to have a good user experience. Lastly, the classification model achieved high performance on local fauna species with the best accuracy of 95.25% among seven species categories. The integration of the front-end and back- end framework is accomplished, and the mobile application performed well and efficient with a bunch of useful features. Besides, a discussion is made to further improve the performance of the classification model and expand the functionality of the mobile application.