School of Infocomm (SOI)

Development of Image Recognition AI to Differentiate Recyclables in Industry

Partner Organisation: GlobalFoundries 

Many recyclable items are sent to landfill or incineration, because they are contaminated with non-recyclables (such as leftover food) which were improperly sorted and disposed of. Although most public recycling campaigns simplify waste categorisation into a few basic rules, it requires a deeper understanding of the types of materials and industrial treatment process to perform the task accurately. This is not common knowledge which should be expected of everyone. 

Key Features: 

• The project uses an artificial demonstration set that can correctly classify household trash against a variety of backgrounds and camera angles.

• It also has an expert system that may ask follow-up questions to verify the results and provide basic cleaning recommendations (such washing the item with soap) for recyclables.
   

Team Members: 
Koh Yik Heng Ryan, Wiz Lie Chu Lok, Syafid Danish Bin Mohamed Harifin, Mohammad Shahrin Idzuan Bin Iswandi

Supervisor:
Mr Ivan Wee

Booklink Transactions Reconciliator

Partner Organisation: Booklink 

Currently for every transaction performed at Booklink RP, the Office of Organisation and Service Excellence (OSE) will receive an email on the transaction details. OSE staff have to manually compile each email into an excel sheet monthly, and compare it to an Excel sheet given by Booklink. This entire process is very demanding and time-consuming. 

Recommended Process: 

This project uses Natural Language Processing (NLP) model to recognise the entities and filter the entities to be populated into the database.

The content in the database will then be populated into a blank Excel sheet to be automatically reconciled with the report given by Booklink, which in turn, eliminates the need for manual compilation.
 

Team Members: 
Ng Surwyn, Tan Bing Jun, Gerard Mah Jie Kang, Daniel Abishak

Supervisor:
Ms Sharmila Kanna

Park Visitors Analytics

Partner Organisation: National Parks Board (NParks) 

During the COVID-19 pandemic, people in Singapore were required to wear face masks when outdoors except during mealtimes and while exercising. However, not everyone adhered to the rules.

Moreover, especially in parks, there were also known problems such as littering of used plastic water bottles and empty canned drinks by visitors.

Main Goal: 

This project aims to enhance the experience of park visitors by detecting whether they are wearing a mask and identifying the location of litter.
 

Team Members: 
Ainsley Chang, Sng Zhi Hao, Tang Sze Chuan Marcus, Leong Jun Wei Dexter

Supervisor:
Mr Seow Khee Wei

Pest Detection Application

Partner Organisation: Huawei International Pte Ltd 

There were growing concerns of infestation as some eateries stopped their pest control services during the COVID-19 pandemic as part of their cost-saving strategy. With less human activity, there would be an increase in pest infestations, which resulted in raising the spectre of another outbreak as Singapore battled against the coronavirus. 

Advantages of Huawei Atlas Robot Car:

• This project incorporates Artificial Intelligence and Optical Recognition (OR) Tools in the application development process to comb areas for signs of pests and to perform pest classification.

• This helps in solving commonly faced urban safety issues and creates a much safer environment for everyone to live in.
   

Team Members: 
Elijah Tan Jia Wei, Ong Yee Fei, Lau Shao Xian

Supervisor:
Mr Jimmy Goh

Logistics Delivery Robot

Partner Organisation: NCS Pte. Ltd

Deliveries have become part and parcel of life for many people. To ensure that the experience is kept to highest standard in terms of convenience and hygiene, contactless mode of delivery has overtaken the traditional one, which involves human interaction. 

Key Features: 

• This project looks at the issues of convenience and hygiene, as well as reduces costly manpower by deploying automated delivery robots.

• Building upon Cloud-Based Technology and coupled with automated programmable delivery robots, the presence of a human being from a goods delivery process is eliminated.
  
  

Team Members: 
Lee Joon Hean Mason, Tan Li Shing, Barry, Soo Shan Yong, Adam Goh Zheng Shan, Mirza Bin Mohamed Aljaru

Supervisors:
Ms Grace Yap and Mr Tan Cheng Kok

RP Chatbot Tutor

Students are not able to contact the lecturers outside school especially on weekends, and when they have questions on their revision or tasks, they are not able to find the answers online as only the lecturers have the answers to their questions.

RP Chatbot Tutor improves user experience in the following ways:

• It allows a student to ask questions and get a reply immediately if the answer is found in the database.

• If the answer is not available in the database, the student can then add in the question into database for the lecturers to answer.

• The project also allows lecturers to manage the questions and answers.

• There is also a dashboard function to allow lecturers to track students’ performance.
  
  

Team Members: 
Lim Kai Peng, Speed, Chu Yanni Sennett Charis, Muhammad Raimi B Hasri, Wee Jun Jie Desmond

Supervisor:
Mr Patrick Wang

RP Chatbot Tutor

RP Chatbot Tutor is a chatbot that is user-friendly and efficient. It answers students’ questions, gives quizzes, and collects questions/feedback from students.

The chatbot came about because lecturers often have to spend a lot of time answering common questions from students. As a result, individual problems from students are not properly addressed. 

Key Features: 

• The chatbot is developed to understand natural language.

• It processes vernaculars or spoken speech the same way humans do. This significantly enhances the user experience.

• Data is collected from the chatbot and displayed on the backend website.

• Only lecturers and staff members have access to the site whereby all modules are managed.
  
  

Team Members: 
Tan Zhi Yin, Ahmad Husaini B Hamsani

Supervisor:
Mr Patrick Wang

Back to top of page

2019

Customer Service Chatbot

Improve Customer Response Time

Main Aim of the Project:

• Our project aims to build a Customer Service Chatbot for NEA that connects to the public. 

• It provides virtual service with 24/7 on demand support to enable immediate responses to customers’ enquiry on operational procedures, payment of fines, illegal dumping, process to handle letter of reminder/advice, work instructions for Singapore Food Agency (SFA), etc. 

• The chatbot relies on Natural Language Processing (NLP) trained using Artificial Intelligence (AI) to understand and correctly interpret customers’ requests.

Final Solution:

• The chatbot system is developed with Google-owned DialogFlow that uses human-computer interaction technologies based on natural language conversations. 

• Further exploration and development are also accomplished using Rasa conversational AI platform. 

• The customer service chatbot is deployed to the cloud through Amazon Web Service (AWS), and it can be communicated via Telegram, a cloud based instant messaging service app on mobile phones and laptops.

Technologies: 

• Amazon Web Services (AWS)
• DialogFlow
• Rasa
• Telegram
   

Team Members: 
Wong Zen Yang, Muhammad B Alfyan Sapwan, Chan Jia Hui, Chia Yun Lydia.

Supervisor:
Mr David Leong

Plant Disease Detection

Modern economies are facing issues regarding food security. Traditional plant-based farms spanning across large areas pose challenges for timely crop monitoring.

With urbanisation increases, farmers are also turning to urban farming. Modern urban farming techniques include vertical farming and rooftop farming. Timely and efficient monitoring is important to ensure crop productivity. 

Plant Drone Flying – An Efficient Solution:

Develop a drone system to detect and scan the health of plants.

• It can fly at a height and scan the leaves of the plants.

• Utilising principles of Deep Learning, it can display the plant's health and send farmers alerts when there are problems with the plants' health. 

Technologies: 

• Drone Flying

• Web Development

• Deep Learning
  
  

Team Members:
Balbin Kristin Clarise Mandap, Reuben Loo Jie Long, Patrick Heng Jun Xiang, Ngieng Min Yi

Supervisor:
Mr Gary Chan

Computer Vision

Real-Time Response 

In sports events, referees are needed to determine the happenings during the event and then perform specific actions based on the key events that they have identified. However, a referee might misjudge some events, hence we want to ensure that all key events are automatically and correctly identified. 

System Requirements:  

• A detection system is required to detect the key events. We are training the models to detect specific objects based on four different algorithms.

• The trained models are used to detect the objects using Tensorflow object detection API.

Custom Trained Detection System Solution: 

A Custom-Trained Detection System will be built to detect 4 different objects in a soccer match. (Penalty Kick, Free-Kick, Throw-In, Corner Kick) using the 4 different algorithms (Single Shot Detectors, YOLO v3, Faster R-CNN and Mask-RCNN). 

With the detection model set in place, it will automatically detect the key events which help to reduce any human errors, increase convenience and accuracy. The output can either be image, video or webcam. 

Technologies: 

• TensorFlow

• NumPy

• Algorithms

• Anaconda virtual environment

• Python
   

Team Members:
Royden See An Jun, Chew Tze Nam, Zhang HongYing, Du Mengxue;

Supervisor:
Mr Zack Toh

Thyroid Nodule Classification

Thyroid cancer is a malignant tumour that occurs in the thyroid gland and is the most common malignant tumour in the endocrine system.

Ultrasound images of thyroid papillary carcinoma are mostly represented by two-dimensional greyscale pixels in lower resolution. It is difficult to distinguish and diagnose due to the complicated internal tissue structure and lack of obvious cancer features.

System Requirements:

• Design and implement a custom Convolutional Neural Network (CNN).

• Perform training and evaluation of well-known models through various hyperparameters tuning techniques.

• Predict and classify thyroid tumour.

AI-based System Solution:

It provides first-level classification and identification of thyroid nodules will help to speed up the analysis, assisting clinicians in determining and locating malignant tumours. 

Technologies: 

• Keras

• Tensorflow

• Python

• Anaconda virtual environment

• Python
  
  

Team Members:
Ng Jia Wei, Tino Chia Jun Rong, Md Haikal Iskandar Bin Osman, Md Haiqal Bin Mohamed Rafiee;

Supervisor:
Mr Seow Khee Wei

Thyroid Nodule Classification

As thyroid cancer is very common worldwide, doctors have to deal with many ultrasound thyroid images. Doctors might not be 100% accurate in their predictions. Usually, the results take a while to be processed too. 

System Requirements:

Design, train and evaluate a Convolutional Neural Network (CNN) and compare its performance against well-known CNN architectures like Inception v3 and VGG16. 

CNN-based Solution:  

With the help of CNN, it will help the doctors to predict the type of thyroid cancer found on the patients effectively. 

Technologies:

• PyCharm 

• Anaconda
   

Team Members:
Phua Guan Wu, Ng Kai Xuan, Han Jian Le, Darren Wong;

Supervisor:
Mr Seow Khee Wei

Back to top of page