DR-NTU (Digital Repository of NTU)
Not a member yet
116018 research outputs found
Sort by
Biological relation recognition from facial features
No full textIn recent years, facial recognition technology has gained notable advancements and
one increasingly explored area in this domain is Facial Kinship verification (FKV).
FKV is still a complex research problem in the computer vision field as it is a binary
classification problem that involves determining biological relationships by comparing
facial features from two distinct individuals’ facial images. Recent psychological
studies have revealed that biologically related individuals exhibit more facial genetic
similarities than those who are unrelated, hence this study explores utilizing deep
learning methods to identify such similarities.
Unlike conventional facial recognition tasks that aim to identify individuals separately,
FKV poses a unique challenge of comparing hereditary facial features between two
distinct individuals. With this challenge in mind, there is a need to explore the use of
deep learning techniques for such tasks. This study proposes a deep learning approach
using a Siamese Neural network together with transfer learning to address the
challenge. The proposed framework aims to improve the feature extraction capabilities
of the Siamese network by employing pre-trained models, thereby optimizing the
ability to identify kinship connections.Bachelor's degre
Software receiver for GNSS positioning
No full textMany applications currently require the positioning capabilities of Global Navigational Satellite systems. Signals emitted from such systems could be easily detected if there was a clear view of the sky and the receiver had a direct line-of-sight to multiple satellites simultaneously. However, for urban canyons, the natural and man-made buildings serve as obstacles for such signals, blocking or reflecting the signals, attenuating them. Such environments then require receivers to have high sensitivity and utilise optimised algorithms to acquire and track these weak signals. In this paper, common signal acquisition techniques used to improve the sensitivity of the receiver are discussed. The implementation of such algorithms to receive Global Positioning System L2 Civil and L5 signals are detailed and simulated results to compare the advantages and disadvantages of each technique.Bachelor's degre
Heterogeneous wireless networks (HWNs) optimisation using machine learning
No full textThis project develops a self-optimizing Heterogeneous Wireless Network (HWN)
using Raspberry Pis, BATMAN-adv, and machine learning techniques to automate
interface selection in the device. The switching rules is based on real-time Quality of
Service (QoS) metrics that consists of latency, packet loss, throughput, and jitter. It is
done to maintain optimal performance under varying network conditions.Bachelor's degre
Investigations of data-driven based control methods for basic power converters
No full textThe widespread deployment of Wireless Power Transfer (WPT) in aerospace, new energy vehicles, and medical devices has heightened demands for precise, rapid modeling of system coupling parameters. The series–series compensated wireless power transfer (SS-WPT) system employs a series-resonance topology on both input and output sides, with a high-frequency inverter driving the resonant network. Magnetic coupling enables efficient energy exchange between primary and secondary coils. The system integrates two series-resonant loops, a high-frequency coupling coil, a rectifier, and an inverter, offering high energy density, a simple structure, and robust resonant performance.
Compared with conventional wired DC–DC converters, SS-WPT systems not only ensure electrical isolation between input and output—substantially boosting reliability—but also simplify resonant frequency design, support soft-switching control (e.g., ZVS/ZCS), and exhibit strong modular scalability. These advantages suit high-frequency, high-efficiency, and contactless power delivery scenarios. In wireless charging, the primary-side inductance L and coupling coefficient k critically influence transmission efficiency and stability, yet are limited by coil dimensions, spatial arrangement, and relative positioning. Conventional analytical methods often struggle with accuracy and adaptability.
To address these challenges, this study introduces a neural-network-based nonlinear prediction model for rapid, accurate estimation of L and k in SS-WPT systems. By training a Multi-layer Feedforward Neural Network (MFNN) on primary-side phase difference, voltage, and current, the model achieves high-precision regression. In data processing, a simulation generates training and validation sets, and normalization plus regularization strengthen generalization and robustness. Experimental findings indicate that this model predicts L and k effectively under diverse conditions, maintaining an average relative error near 7%. It also demonstrates strong scalability, making it applicable to more complex wireless charging structures. Consequently, the proposed approach provides an efficient modeling tool and technical support for system optimization and real-time control in WPT applications.Bachelor's degre
LOOPod - looped transit pod for Singapore tourism
No full textThis project introduces an autonomous electric vehicle (AEV), LOOPod (Looped Transit Pod), designed to enhance the visitor experience at tourist attractions in Singapore. It offers a premium, private space where guests can escape the heat, rain, noise, and crowds. Whether enjoying scenic views with a glass of wine, stopping for photos, or simply unwinding, this AEV makes every visit more relaxing and memorable.
Unlike self-driving cars on public roads, LOOPod operates safely within designated areas at low speeds. This eliminates the need for complex AI to navigate traffic, making it safer and more cost-effective.
Integrating LOOPod into major attractions can help draw more visitors and increase revenue. It offers a fresh, enjoyable way to explore, especially for those who value comfort and convenience. Families, in particular, can provide a better experience for children and elderly members, ensuring a more relaxing and stress-free visit.Bachelor's degre
Design of large bandwidth phase locked loop circuit
No full textA wide-bandwidth Phase-Locked Loop (PLL) has extensive application prospects in modern communication systems, radar and navigation systems, precision measurement instruments, and consumer electronics. It holds significant research value in enhancing system response speed, improving noise suppression capabilities, meeting diverse frequency demands, and driving innovations in circuit design. Its broad application not only significantly enhances the performance of modern communication and electronic systems but also promotes the development and innovation of related technologies.
This dissertation designs a wide-bandwidth PLL, with each module of the PLL designed individually, followed by system-wide simulation verification, based on smic18mmrf PDK. The dissertation first explains the working principles of each module, introduces common structures of each module, analyzes their mathematical models, and then details the design of each module with simulation evaluations. The design and simulation of each module are completed, and simulation results indicate that the performance of each module meets the required standards.
At the end of the dissertation, layout design considerations are briefly summarized, the overall layout is presented, and the final simulation results are shown. The simulation results demonstrate that the designed PLL system can achieve a frequency range coverage of 8~12 GHz, with an output clock peak-to-peak jitter of less than 15 ps.Master's degre
Math field-specific ability beliefs and career aspirations in adolescents: mediating effects of interest and self-efficacy
No full textThe mechanisms through which math FABs affect math job aspirations
are largely unknown.
Few studies have been conducted on early-adolescence, a crucial
developmental stage for constructing career identity
Mechanical properties of sandwich structure with entangled metallic wire material composites under low-velocity impact
No full textSandwich structures are highly versatile and have good natural strength. Therefore,
sandwich structures are commonly used in engineering structures in a multitude of
industries. In addition, to allow for sandwich structures to be used in an even wider
array of industries and applications, experiments and studies can be conducted to
verify the applicability of sandwich structures with different core and face plate
materials. In recent years, there have been a growing interest in using composite
materials as core materials for sandwich structures. Using composite materials allows
for the sandwich structure to have different behaviors and reactions to different
environments, loading conditions and applications.
This research aims to study the behavior and reaction of a sandwich structure with
entangled metallic wire material boned with silicone rubber as a composite core
material, under low velocity impact load testing. The low velocity impact load tests
was conducted by dropping a weight onto the sandwich structure and recoding the
results. Placing more emphasis on comparing the sandwich structure’s energy
absorption capabilities and response to dynamic loading. The experiment showed
that the sandwich structure was able to resist puncturing when subjected to up to
100J of impact energy, indicating good energy absorption capabilities. This
experiment validates and allows for the specific material; entangled metallic wire
with silicone rubber composite as a novel material for a sandwich structure core.Bachelor's degre
Assessing Malay speakers’ receptiveness to Indonesian clipped compound neologisms
No full textBahasa Melayu’s ‘Malay language’ usage in Singapore has declined over the years but Bahasa Indonesia ‘Indonesian language’ remains relevant and widely spoken in Indonesia. A driving
factor for this phenomenon could be contributed by the youth language in Indonesia, Bahasa Gaul, that is enriched with neologisms such as clipped compound words that shapes the youth culture in Indonesia and increases the longevity of Bahasa Indonesia. Meanwhile, Bahasa Melayu’s usage has declined over the years, facing enormous threats from the dominance of English. Thus, this study explores the reception of Bahasa Melayu speakers to Bahasa Gaul clipped compound neologisms, seeking the possibility of the neologisms enhancing the use of
Bahasa Melayu in Singapore. Findings indicate that while participants could identify and decompose the neologisms with exposure, overall receptiveness is between against and in favour
where resistance and concerns are highlighted. To add on, this study discovered the absence of youth language literature of Bahasa Melayu in Singapore, highlighting the gap for further research in this area. These results contribute to discussions on linguistic innovation, language vitality, and the role of these informal linguistic features shaping linguistic change of Bahasa Melayu in Singapore.Bachelor's degre
Developing a demonstration system of multi-modal search
No full textVast amounts of data and knowledge are created daily, yet most remain undiscovered
with ineffective search systems. Time is wasted in fruitless search and in recreating
existing knowledge that could not be found. Modern retrieval systems either depend on
rigid text-based indexing or resort to performance-degrading ways of bridging multiple
modalities, such as converting all documents to a single modality or using multimodal
models.
This report proposes a novel multi-agent multimodal retrieval framework, designed
with a focus on retrieval accuracy and extensibility. When documents are uploaded,
primitive elements such as texts and images are extracted and indexed in separate
vector database indexes using element-specific embedding models. To serve queries,
each embedding model will be used to retrieve the top-k most similar elements from
the respective indexes. The elements are re-ranked by their respective element-specific
re-rankers, and the top-n chunks are yielded for each element type.
A demonstration of such a retrieval system has been built, with design considerations
and implementation details thoroughly discussed in this paper. The underlying architecture uses a hybrid of event-driven architecture and request-response model to mimic
a practical implementation of such a system in the real world.
This framework is recommended for institutions to build powerful and extensive search
systems, with the potential to massively boost productivity through effective, streamlined and simplified search.Bachelor's degre