<?xml version="1.0" encoding="UTF-8"?>
<feed xmlns="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
<title>Ph.D/Master Thesis &amp; Dissertation</title>
<link href="https://onlineresource.ucsy.edu.mm/handle/123456789/2229" rel="alternate"/>
<subtitle/>
<id>https://onlineresource.ucsy.edu.mm/handle/123456789/2229</id>
<updated>2026-11-17T15:13:52Z</updated>
<dc:date>2026-11-17T15:13:52Z</dc:date>
<entry>
<title>A Spatial-Temporal Attention Framework for Human Fall Detection in RGB Videos</title>
<link href="https://onlineresource.ucsy.edu.mm/handle/123456789/2843" rel="alternate"/>
<author>
<name>Win, Aye Mya Mya</name>
</author>
<id>https://onlineresource.ucsy.edu.mm/handle/123456789/2843</id>
<updated>2026-11-17T08:15:26Z</updated>
<published>2026-11-01T00:00:00Z</published>
<summary type="text">A Spatial-Temporal Attention Framework for Human Fall Detection in RGB Videos
Win, Aye Mya Mya
Timely and accurate human fall detection is essential for enhancing the safety&#13;
of older adults and individuals who require assistance. In video-based human fall&#13;
detection, variations in movement speed and motion patterns across different activities,&#13;
as well as changes in body posture and movement throughout an action, can be made&#13;
accurate activity recognition challenging. In addition, differences in camera placement,&#13;
recording distance, and viewing angles can lead to variations in the appearance and&#13;
posture of a person across video frames. Therefore, reliable fall detection requires the&#13;
effective analysis of both spatial features related to human appearance and temporal&#13;
patterns that represent movement over time. To address these requirements, this&#13;
research proposes a deep learning-based video classification model that combines a&#13;
Convolutional Neural Network (CNN), a Temporal Convolutional Network (TCN), and&#13;
an Attention Mechanism.&#13;
The proposed model extracts spatial features related to human posture and&#13;
visual appearance from RGB video frames using CNN. The resulting feature sequences&#13;
are then processed by TCN to learn movement patterns and temporal dependencies. In&#13;
addition, the attention mechanism helps the model emphasize informative features and&#13;
temporal segments relevant to fall classification. By integrating spatial and temporal&#13;
representations, the model aims to distinguish fall events from post-fall recovery and&#13;
other non-fall activities.&#13;
The Multiple Cameras Fall Dataset (MCFD) is used for model training and&#13;
evaluation. The performance of the proposed CNN-TCN-Attention model was&#13;
compared with that of the CNN-LSTM model for fall detection. The result showed that&#13;
the CNN-TCN-Attention model achieved a test accuracy of 84.88%, outperforming the&#13;
CNN-LSTM model, which obtained 82.56% under the same experimental settings.
</summary>
<dc:date>2026-11-01T00:00:00Z</dc:date>
</entry>
<entry>
<title>Multi-Object Tracking with an Improved DeepSORT Framework and YOLOv9</title>
<link href="https://onlineresource.ucsy.edu.mm/handle/123456789/2842" rel="alternate"/>
<author>
<name>Aung, Phway Phway</name>
</author>
<id>https://onlineresource.ucsy.edu.mm/handle/123456789/2842</id>
<updated>2026-11-17T04:25:27Z</updated>
<published>2026-11-01T00:00:00Z</published>
<summary type="text">Multi-Object Tracking with an Improved DeepSORT Framework and YOLOv9
Aung, Phway Phway
Since information as the demand for automated scene understanding and&#13;
persistent visual monitoring increased, Multi-Object Tracking (MOT) has become an&#13;
essential component in various real-life applications, like smart video surveillance,&#13;
traffic monitoring, robotics, sports analytics, customer behavior analysis and drone&#13;
control. Nevertheless, MOT remains challenging due to complex visual conditions such&#13;
as heavy occlusion, frequent identity switching, variable illumination, and inter-object&#13;
similarity. The proposed system outperforms a novel MOT framework integrating the&#13;
YOLOv9 multi-object detector with a DeepSORT algorithm based on the Generalized&#13;
Intersection over Union (GIoU) metric and an enhanced matched and filtering&#13;
algorithm to improve detection accuracy, reduce ID switching and occlusion&#13;
challenges. This integration aims to effectively tackle the critical challenges of ID&#13;
switching and occlusion, thereby enhancing multi-object tracking accuracy on the MOT&#13;
16 datasets. The suggested multi-object tracking system applies a refined matched filter&#13;
approach featuring two major improvements: a class-based validation step ensuring&#13;
detection consistency and a re-identification memory component preserving unmatched&#13;
tracks for possible restoration. The proposed system results are compared with the&#13;
DeepSORT algorithm, which uses standard Intersection over Union (IoU).&#13;
Measurement results show the proposed conversion method greatly improves tracking&#13;
performance, offering better accuracy and identity preservation than the standard IoU&#13;
system. The experiment demonstrated that the improved system keeps object identities&#13;
consistent during occlusions, lowers ID switches, and boosts overall tracking accuracy&#13;
over the DeepSORT baseline.
</summary>
<dc:date>2026-11-01T00:00:00Z</dc:date>
</entry>
<entry>
<title>Inflow Prediction of Lower Paunglaung Dam for Water Resource Management</title>
<link href="https://onlineresource.ucsy.edu.mm/handle/123456789/2841" rel="alternate"/>
<author>
<name>Oo, Zin May</name>
</author>
<id>https://onlineresource.ucsy.edu.mm/handle/123456789/2841</id>
<updated>2026-11-17T04:20:30Z</updated>
<published>2026-11-01T00:00:00Z</published>
<summary type="text">Inflow Prediction of Lower Paunglaung Dam for Water Resource Management
Oo, Zin May
Dams are essential to Myanmar's energy and water supply sectors because they&#13;
support irrigation, flood control, hydropower generation, and the transport of water for&#13;
homes and farms. To operate these reservoirs efficiently, an accurate forecasting of&#13;
water availability in the future is necessary. Accurate dam inflow forecast is also crucial&#13;
for reservoir operation, flood mitigation, hydropower generation and sustainable water&#13;
resource management.&#13;
This system aimed to implement an effective deep learning daily inflow&#13;
forecasting model at Lower Paunglaung Dam, Myanmar, using the catchment area&#13;
rainfall and relevant hydrological data. Daily time-series historical data from 1 January&#13;
2015 to October 15 2026 were used to develop and evaluate the models. Three types of&#13;
experiments were conducted to assess the contribution of Long Short-Term Memory&#13;
(LSTM) model and the performance of the proposed model. The first experiment used&#13;
conventional LSTM and Gated Recurrent Unit (GRU) models without rainfall&#13;
information. The second experiment was conducted to evaluate the effect of rainfall as&#13;
additional information for inflow prediction and the third experiment utilized the&#13;
proposed Rainfall-Guided Forget Gate LSTM (RGF-LSTM) model, where the rainfall&#13;
information is integrated directly into the forget gate of a standard LSTM cell. The&#13;
proposed system aims to predict the inflow of Lower Paunglaung Dam over a seven-&#13;
day forecasting period using historical inflow, rainfall, and other relevant hydrological&#13;
data. The predicted inflow will support water resource management for hydropower&#13;
generation by providing information for reservoir water management and electricity&#13;
generation planning. The system is designed to help dam operators make informed&#13;
decisions regarding water utilization for hydropower generation.&#13;
The models' performance was examined using Root Mean Square Error&#13;
(RMSE), Mean Absolute Error (MAE), Coefficient of Determination (R2). The&#13;
experimental results showed that the RGF-LSTM model obtained the best overall&#13;
performance for the daily dam inflow forecasting.
</summary>
<dc:date>2026-11-01T00:00:00Z</dc:date>
</entry>
<entry>
<title>Security Analysis of a Modified AES Algorithm Compared with Existing Symmetric Block Ciphers</title>
<link href="https://onlineresource.ucsy.edu.mm/handle/123456789/2840" rel="alternate"/>
<author>
<name>Oo, Thaw Thaw May</name>
</author>
<id>https://onlineresource.ucsy.edu.mm/handle/123456789/2840</id>
<updated>2026-11-17T04:20:28Z</updated>
<published>2026-11-01T00:00:00Z</published>
<summary type="text">Security Analysis of a Modified AES Algorithm Compared with Existing Symmetric Block Ciphers
Oo, Thaw Thaw May
Secure file sharing and concurrent access management are essential&#13;
requirements in collaborative environments, particularly when multiple users access&#13;
and modify shared files. This research proposes a group-based collaborative filesharing system that integrates a Modified Advanced Encryption Standard (MAES)&#13;
algorithm with a Timestamp-Based Concurrency Control (TBCC) mechanism. The&#13;
proposed MAES algorithm retains the fundamental structure of AES while&#13;
incorporating a key-dependent Dynamic S-Box and a key- and block-dependent&#13;
Dynamic ShiftRows transformation. The TBCC mechanism enables multiple users to&#13;
access shared files in ReadOnly mode while restricting editing access to one user at a&#13;
time. The cryptographic characteristics and processing performance of MAES are&#13;
experimentally evaluated. Processing performance is assessed in terms of encryption&#13;
time, decryption time, CPU utilization, peak additional memory usage, and&#13;
throughput. Ciphertext characteristics are examined using the avalanche effect,&#13;
ciphertext entropy, and key sensitivity, with Twofish and Blowfish serving as&#13;
comparative algorithms. Furthermore, the Dynamic S-Box is evaluated based on the&#13;
Strict Avalanche Criterion (SAC), Bit Independence Criterion with respect to&#13;
nonlinearity (BIC-NL), algebraic degree, and invertibility. Experimental results&#13;
indicate that the Dynamic S-Box achieves an average SAC value of 0.5039, a BIC-NL&#13;
value of 103.50, and an algebraic degree of 7.00. Its invertibility is successfully&#13;
verified across all 256 possible input values. For MAES, the measured avalanche&#13;
effect is 49.35%, ciphertext entropy is 7.985127 bits per byte, and key sensitivity is&#13;
50.0054%. These results are compared with those obtained from Twofish and&#13;
Blowfish to examine the cryptographic characteristics of the evaluated algorithms.&#13;
Overall, this research provides an experimental assessment of the cryptographic&#13;
characteristics and processing performance of MAES, together with the functional&#13;
operation of the integrated collaborative file-sharing and concurrency-control&#13;
mechanisms.
</summary>
<dc:date>2026-11-01T00:00:00Z</dc:date>
</entry>
</feed>
