Visual analytics using deep learning : fine-grained image recognition of fauna species in Singapore
In a fast-paced urbanized world, there is a growing less connection between the human and the natural environment. To encourage communities to bond over and with nature, we proposed a rich-functionality mobile application using fine-grained image classification algorithm that can recognize the fauna...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2021
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/149906 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | In a fast-paced urbanized world, there is a growing less connection between the human and the natural environment. To encourage communities to bond over and with nature, we proposed a rich-functionality mobile application using fine-grained image classification algorithm that can recognize the fauna species in Singapore near instance. This is a full-stack mobile application development project. It’s divided into three parts which are deep learning model development, front-end and back-end development.
First of all, we generated a new fauna image dataset and classified it into 7 super categories which include bird, butterfly, dragonfly, amphibian, reptile, mammal, and fresh-water fish. The data is crawled by using Flickr API, followed by data cleaning, data annotation, and data pre-processing. Furthermore, the state-of-the-art fine-grained image classification algorithm Attentive Pairwise Network (API-Net) is used to train the model with PyTorch deep learning framework.
For the development of the mobile application, React Native framework is used to designed and created the user interface, and then the classification model is deployed on the AWS Lambda to host the web service, furthermore, MangoDB is used as a database to store user’s information and fauna data.
Lastly, both the front-end and back-end are successfully integrated, and the classification models have a high performance on recognizing local fauna species with the best accuracy of 95.25%. |
|---|