border border

Corabot: An Autonomous System for Coral Reef Monitoring

Team members

Ong Wei Song (EPD), Lewin Sim Le Wei (EPD), Tan Sok Ming Jamie (EPD), Ong Jia En (EPD), Yee Wang Sui (ESD), Chan Jie Lin (ISTD), Koh Xian Ming (ISTD)


Nagarajan Raghavan, Keegan Kang, Cyrille Pierre Joseph Jegourel

Writing Instructors:

Nurul Wahidah Binte Mohd Tambee

Teaching Assistant:

Ataman Cem

Problem: Coral Reef Monitoring in Singapore

Coral reef monitoring is a highly laborious 6-step process. The problem is further compounded by the non-ideal water conditions in Singapore. Singapore's shallow and murky waters with strong undercurrents and high shipping traffic hinder the use of commercial off-the-shelf reef monitoring robots in local waters.


before only own size

Solution: Corabot

after only own size

An automated and economical end-to-end solution was designed to solve the problem. An autonomous underwater vehicle (AUV) collects image data and transfers it to the cloud. Artificial intelligence then colour corrects, classifies and segments coral species before displaying results on an intuitive graphical user interface (GUI).


1. Underwater Deployment

To eliminate risks associated with diving, an AUV is deployed to collect coral data on behalf of marine biologists. It was designed to be:


i. Portable

Weighing about 12kg, the AUV is lighter than most commercially available reef monitoring robots.


ii. Easy and safe to deploy (video 1)

A foldable handle was designed for the safety of the operator. With additional height, the operator need not bend over the boat excessively, thereby reducing the risk of falling overboard. He can also deploy the AUV single-handedly, giving him one free hand to hold onto the boat.


iii. Maneuverable (video 2)

Equipped with five thrusters, the AUV has five degrees of freedom: surge, heave, roll, pitch and yaw. Coupled with finetuned controllers, the AUV can maintain its path despite strong undercurrents and remain stable for the collection of high quality image data.


iv. Easy to operate and track (image 1)

Remote control of the AUV using a radio switch and a GUI allows the operator to turn on the AUV and track its battery levels, mission status and location remotely.


v. Long battery life (video 1)

A dual battery system reduces the time-consuming task of constantly recharging the AUV. An internal battery powers the electronics within the main hull. External batteries, which can be easily replaced, power the high energy consumption thrusters.


2. Remote Data Collection

During deployment, the AUV fulfills two main functions:


i. Collect high quality images

Equipped with a 12MP downward-facing camera, the AUV collects high-resolution coral images to ensure data quality.


ii. Comprehensive reef scannning

A lawn-mowing algorithm allows the AUV to scan an area of 400mat each reef site.

lawnmowing png


3. Automated Data Processing

After the AUV surfaces, the image data is transferred to the cloud where artificial intelligence is employed to automate the data processing in a matter of seconds, which would otherwise take up to three weeks if the researcher does it manually. This process consists of 3 stages:


i. Colour Correction (image 2)

Underwater images often suffer from colour degradation due to light absorption and scattering in water. State-of-the-art UGAN algorithm was used to show the true colour of corals, with higher contrast, less blur and less colour distortion. This helps to improve human understanding of the underwater scene.


ii. Coral Species Prediction (image 3)

  • Corals can be difficult to identify due to their varying shapes and colours, even within a species, resulting in the need for references to coral identification guides. The species prediction algorithm based on the DenseNet architecture provides a 96% confidence that the correct species is in the top 3 predictions, helping to narrow down the options. 


iii. Coral Segmentation (image 4)

Manual tracing of a coral's outline is required to determine its size. It is a painstaking process done for every image taken during a coral monitoring dive. The coral segmentation algorithm with the famous UNet architecture and EfficientNet-B6 encoder serves to automatically generate the coral outline with a 95% F1 score, saving time and effort. 


4. Quick Data Visualisation

The results of the algorithms are displayed on a GUI. As the models are not 100% accurate, the GUI has an element of human computer interaction where the user can modify and verify the results. This feedback is used for further finetuning and improvements of the models.


gui 2

"With the AUV covering more area across different sites, more information can be retrieved within a shorter period. With lesser money spent on boat charter, project funds for charters can be relocated to other resources or to expand the project scope. The AUV can also help researchers to assess the underwater conditions of unexplored or deeper sites."

- Toh Tai Chong, Senior Lecturer, College of Alice & Peter Tan, Tropical Marine Science Institute

"Cool! I am very impressed already. This is what is needed on the public domain! A lot of scientists are still using the old-school methods.

The GUI can potentially help to save a lot of time, especially in measuring coral areas and preliminary identifying the corals, so we can spend more time on data analysis and writing reports.

It can also be used as a tool to train researchers and citizen scientists in coral reef monitoring."

- Sam Shu Qin, Research Assistant, Tropical Marine Science Institute


Summary of Features


Meet the Team



In Collaboration With:                            Supported By:


DSO National Laboratories Logo


student Ong Wei Song Engineering Product Development
student Lewin Sim Le Wei Engineering Product Development
student Tan Sok Ming Jamie Engineering Product Development
student Ong Jia En Engineering Product Development
student Yee Wang Sui Engineering Systems and Design
student Chan Jie Lin Information Systems Technology and Design
student Koh Xian Ming Information Systems Technology and Design
border border