Post graduate dissertations (Theses) of Institute of Information and Communication Technology (IICT) http://lib.buet.ac.bd;localhosthttp://:8080/xmlui/handle/123456789/661 2026-05-26T03:39:41Z 2026-05-26T03:39:41Z Sheezanul Hassan http://lib.buet.ac.bd;localhosthttp://:8080/xmlui/handle/123456789/7323 2026-05-19T03:53:10Z 2024-10-19T00:00:00Z

An automated walking guide to assist way-finding and situation awareness for the visually impaired Liakot Ali, Dr. Md.; Sheezanul Hassan; 1018312010; 006.424/SHE/2024 Safe and independent mobility is one of the major daily challenges faced by the visually impaired. To navigate a new area safely, they need to know the location of obstacles and other things in their path. They struggle with object detection and obstacle avoidance, making it challenging to navigate new or unfamiliar situations and be aware of obstacles and their relative positions. However, establishing secure and safe mobility and pathfinding for the visually impaired is a critical issue in their life that must be solved accurately and efficiently. Recognizing currency is another severe problem for them because different notes in our country have similar colors, surfaces and sizes causing major problems for the visually impaired. Object recognition alone may not be sufficient to assist visually impaired individuals effectively. Incorporating lateral position identification can provide users with a sense of spatial orientation within their environment, enabling them to navigate paths toward recognized objects more accurately. In this thesis, a system is proposed to assist visually impaired individuals by identifying both navigation objects and their corresponding lateral positions, and recognizing Bangladeshi currency. This system aims to serve as a comprehensive walking guide, offering benefits for both indoor and outdoor navigation. By enhancing spatial awareness and providing critical information about their surroundings, this system will enable visually impaired users to make informed decisions regarding their movements and actions; thereby, improving their independence and quality of life. The system utilizes EfficientNet, a Convolutional Neural Network architecture for Machine Learning to design a model for currency and navigation objects that can recognize fifteen classes of objects for assisting the visually impaired. The architecture also applies computational logic to the model to identify lateral positions of the navigation objects; and thus conceptualizing a system that can act as a walking guide for visually impaired people aiding navigation and awareness. The model proposed in the system has been evaluated with real-world objects to assess the performance of the proposed method. The experimental analysis demonstrates that the system achieves notable Accuracy, Precision, Recall, and F1 scores, highlighting its prospective relevance and effectiveness in the field.

2024-10-19T00:00:00Z Shakil Ahammad http://lib.buet.ac.bd;localhosthttp://:8080/xmlui/handle/123456789/7313 2026-05-04T03:52:33Z 2025-02-19T00:00:00Z

Generation of novel mashup data and detection of categorical web attacks Asiful Mustafa, Dr. Hossen; Shakil Ahammad; 0417312012; 005.7/SHA/2025 Information security is a dynamic combination of technology between what we have and whatever we can. The study of information security is all about data because of the polymorphic nature of cyber-attacking signatures. Utilizing critical data, such as at- tacking signatures, is a primitive component of designing green cyberspace. Designing sustainable cyberspace by planning different critical systems utilizing machine learn- ing and Red Teaming concepts is important to protecting digital assets in the globally connected world. A heterogeneous source of IOC data can be an excellent source for designing a minified version of SIEM, and this research has made a paradigm about m-SEIM components. To design this m-SEIM, a few premium enterprise Dynamic Application Security Testing (DAST) tools, like Acunetix-360, Qualys WAS module, Burpsuite Enterprise edition, Nessus, etc, have been used to collect web-attack signa- tures. These tools are considered a great source of attacks in the cyber security in- dustry. As a plan for the cost-effective cyberspace design concept, this work describes how the data of the attacking side can be re-engineered using machine learning tech- niques. From these, heterogeneous sources of attacking signatures are considered Novel Mashup Data (NMD). As part of the optimistic model-finding approach of this research on the novel mashup cyber-attacking data, the success ratio of the attacking and be- nign data is 99.67% on CNN. In comparison, the Lagrangian Support Vector Machine (LSVM), Logistic Regression, and Recurrent Neural Network (RNN) models success ratio has been observed respectively 99.63%, 99.38% and 99.37%. In this work, four types of web attacks, e.g., Path-Traversal, OS Commanding, SQLi, and XSS, are con- sidered to be detected through machine learning. A supervised machine learning model Convolutional Neural Network (CNN) has been trained, and the success ratio has been observed at 98.55% on the collected data. However, the collected data is not equally sampled due to the limitation of the cyber-attacking signatures. The trained model is deployed with the web server to calculate the run-time threats of the operand message. Open-source Kali Linux tools are utilized to check the model’s performance for data poisoning attack detection and classification purposes. In this phase of the performance measures, we have observed the success of Path Traversal, OS Commanding, SQLi, and XSS attack classification success, respectively 91.78%, 53.52%, 95.2%, and 89.18%.

2025-02-19T00:00:00Z Nasima Islam Bithi http://lib.buet.ac.bd;localhosthttp://:8080/xmlui/handle/123456789/7310 2026-04-08T06:35:28Z 2025-04-07T00:00:00Z

An efficient CNN-based regression model for noise prediction in color images Miah, Dr. Md Jarez; Nasima Islam Bithi; 0421312032; 006.42/NAS/2025 The presence of noise in digital images remains a critical barrier to high-fidelity visual representation, compromising both perceptual quality and the accuracy of downstream image analysis. Among various noise types, Gaussian and Poisson noise stemming from electronic sensor imperfections and photon-counting variability, respectively pose significant challenges. Accurate estimation of these noise levels is essential for improving computer vision tasks such as denoising, super-resolution, segmentation, and object detection. However, existing approaches often struggle under high-noise conditions, mixed-noise scenarios, and color image contexts. To address these limitations, we introduce two deep regression models: NoiseNet and NoiseNetV2, designed with noise-specific feature extraction and attention mechanisms to enhance performance across diverse noise types and datasets. Evaluated on datasets including Flickr30k, COCO, CelebA, and DIV2K, the proposed NoiseNet model achieved state-of-the-art results. For Gaussian noise, it recorded the Mean Absolute Error (MAE) of 0.0038, a Root Mean Squared Error (RMSE) of 0.0052, and an R² score of 0.9910, outperforming the best-performing baseline DenseNet121 by margins of ~2.7× lower MAE, ~2.5× lower RMSE, and ~1.05× higher R² score. For Poisson noise, NoiseNet achieved an MAE of 0.0547, RMSE of 0.0763, and R² of 0.9922, representing improvements of ~ 4× lower MAE, ~3.7× lower RMSE, and ~1.1× higher R² score compared to the best-performing baseline ResNet50. The model also outperformed classical techniques like BM3D and Scikit-learn in both low- and high-noise scenarios. Furthermore, NoiseNetV2, an enhanced variant incorporating an attention mechanism, delivered even stronger results under mixed noise conditions. For Gaussian noise, it reached an MAE of 0.0078, R² of 0.9682, and RMSE of 0.0103; for Poisson noise, it obtained an MAE of 0.2759, R² of 0.8106, and RMSE of 0.3763, outperforming both NoiseNet and all other baseline models for mixed noise. These findings confirm the proposed models’ strong generalization and practical utility in fields such as medical imaging, autonomous driving, remote sensing, and surveillance, enabling precise noise estimation for advanced image restoration and real-time vision systems.

2025-04-07T00:00:00Z Zobair Raihan, Md. http://lib.buet.ac.bd;localhosthttp://:8080/xmlui/handle/123456789/7240 2026-01-21T04:53:07Z 2025-03-10T00:00:00Z

Comprehensive ensemble-based machine learning approach for DDoS detection to enhance security in software-defined network Saiful Islam, Dr. Md.; Zobair Raihan, Md.; 1017312030; 006.31/ZOB/2025 Software-Defined Networking (SDN) has emerged as a key solution for meeting the expanding demands of IT and online computing services. Its ability to offer flexible network management and cost-efficient operations has made it a preferred choice for businesses across various industries. However, SDN environments are highly susceptible to security threats, particularly Distributed Denial of Service (DDoS) attacks, which can severely impact network performance and economic stability. Addressing these security concerns is crucial to ensuring the reliability and resilience of SDN-based infrastructures. This research proposes an effective solution to detect DDoS attacks using ensemble-based machine learning. Ensemble models excel at analyzing network traffic, identifying hidden patterns, and distinguishing between benign and malicious packets. This capability helps prevent unnecessary costs incurred by users due to DDoS attacks. However, the use of black-box machine learning models in DDoS detection raises concerns about false positives (legitimate packet rejection) and false negatives (malicious packet acceptance). To address this issue, explainable AI techniques SHAP and LIME have been employed to enhance model interpretability, providing insights into the decision-making process. The CIC-DDoS-2019 dataset was utilized for model development and evaluation. Results demonstrate that the ensemble-based model outperforms non-ensemble models on the dataset. With an outstanding accuracy of 0.9999 for binary classification and 0.9627 for multi-class classification, the XGB model outperformed the others in both tests. The model's choices were interpreted using SHAP analysis, which highlighted the most significant aspects. Top contributing features in the multi-class scenario were 'Min Packet Length,' 'Fwd Packet Length Min,' 'Flow Bytes/s,' and 'Inbound.' 'Packet Length Std,' 'Destination Port,' 'Bwd Packet Length Max,' 'Bwd Packet Length Mean,' and 'Fwd PSH Flags' turned out to be the most important feature influencing the model's predictions for binary classification. Additionally, SHAP and LIME were used to explain individual model predictions, ensuring transparency in the decision-making process. This research highlights the effectiveness of ensemble-based models in DDoS detection and the importance of explainable AI in improving trust and reliability in cybersecurity applications.

2025-03-10T00:00:00Z