Page 56 - Undergraduate Research Opportunities Program Proceedings 2017-18

School of Engineering
Department of Civil & Environmental Engineering

Department of Civil & Environmental Engineering

Development of a Red Tide Early Warning System for Fisheries Management

Supervisor: CHAN Shu Ning / CIVL Co-Supervisor: LEE Joseph Hun-wei / CIVL
Student: KAM Chun Hao / EVMT Course: UROP1100, Spring

A red tide warning system benefits local fisheries by allowing early preparatory management to minimize fish loss. To
facilitate forming predictions for red tides, this research project studied the real time measurements from a continuous
monitoring site at Yung Shue Au. The measurement data include water temperature, salinity, dissolved oxygen and
chlorophyll concentration. It was found that red tides generally occur at moderate to high water temperature (15 °C ~
30 °C) and high salinity (30 ppt). Dissolved oxygen depletes as red tides develop. However, the real time data could not
serve as a predictive guide for red tides due to the frequent data gaps and suspected measurement errors.
Improvements in the robustness of real time data sensors should be made before developing prediction models for red
tides.

Image-based Structural Analysis Using Deep Convolutional Neural Network

Supervisor: CHANG Chih-chen / CIVL
Student: LEUNG Ho Ming / CIVL Course: UROP1100, Summer

Hand-drawn structural diagrams are the common and direct tool for civil engineers to express the blueprint of their
design. Notwithstanding the high demand, a convenient software that can perform structural analysis by hand-drawn
diagrams is yet to be developed. The conventional way of performing structural analysis by using the hand calculation is
timeconsuming. Using finite element software to solve the structural system could be a possible way but learning those
complicated software could be difficult to the students and new engineers. Hence, this project proposes Image-based
Structural Analysis (ISA), which can directly generate the structural response by recognizing the hand-drawn structural
images. Up to now, selective search and deep Convolutional Neural Network (CNN) have been designed for object
detection. Extending the works from previous UROP and FYP, this paper extends the preparation of 2-D frame structure.
The details of preparation of the 2-D frame structure problem and the image labelling work are discussed.

Supervisor: CHANG Chih-chen / CIVL
Student: NG Zhi Yong Ignavier / COSC Course: UROP2100, Fall

Hand-drawn schematic drawings of a structural system are often adopted by civil engineers and engineers-to-be to
study the response of the structure under a set of external loading. Modern finite Element Model (FEM) softwares allow
users to analyze the response of the system quickly. However, it is often difficult to use FEM software due to having to
learn new methods, which may require even more effort to input data. The aim for the UROP project is then to create a
system to have the best of both worlds, which would be to directly extract information from hand-drawn diagrams and
automatically reconstruct the model in the FEM software. With the recent advances of machine learning, this study
proposes a novel deep Convolutional Neural Networks (CNNs) for image recognition of the structural components in the
diagram. The study first reviews the work done by previous UROP students, followed by pipelining of software
engineering practices like code refactoring and fine-tuning of machine learning system like multiple predictors model.

46