http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/16555 Mon, 27 Apr 2026 00:31:20 GMT 2026-04-27T00:31:20Z http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/16741 Correlation Between Air Quality And Respiratory Health Saha, Chinmoy This thesis analyzes the relationship between air quality and respiratory health that develops a predictive framework to inform public health planning. Measurements of key pollutants (PM2.5, PM10, NO2, SO2, O3) and meteorological variables (temperature, humidity, wind) were combined with respiratory health outcomes to quantify associations and forecast short‐term risk. After data preparation and normalization, an artificial neural network (ANN) was trained for regression (estimating health burden) and classification (assigning risk categories) and evaluated using standard metrics. The analysis indicates that elevated concentrations of fine particulates and nitrogen dioxide are consistently associated with increased respiratory morbidity, while meteorological conditions modify exposure– response patterns and improve predictive performance. The resulting model demonstrates practical utility for early warnings, clinical preparedness, and targeted mitigation in high‐risk locations. Overall, integrating environmental monitoring with machine learning can shift practice from reactive management to proactive prevention. Future work should evaluate generalizability across regions, incorporate additional health endpoints and socioeconomic context, and assess operational deployment and cost‐effectiveness under changing climate conditions. Thesis Sun, 14 Sep 2025 00:00:00 GMT http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/16741 2025-09-14T00:00:00Z http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/16740 Semi-Supervised Ultrasound Framework with Prototype Regularization for Five-Stage Liver Fibrosis Assessment Shakil, Md. Shahriar Liver fibrosis represents a gradual replacement of normal hepatic parenchyma with scar tissue, and accurate staging (F0–F4) is essential for guiding surveillance, therapeutic decisions, and specialist referral. Although biopsy remains the historical gold standard, it is invasive, costly, and prone to sampling variability. Ultrasound (US), by contrast, is non- invasive, broadly accessible, and suitable for repeated monitoring in chronic liver disease; however, intrinsic challenges such as depth-dependent attenuation, speckle noise, and inter-scanner heterogeneity complicate automated staging. In this study, developed an explainable semi-supervised learning framework for five-stage fibrosis classification using heterogeneous US data from tertiary-care hospitals, capturing real-world class imbalance with predominance of F0 and F4 cases. The framework integrates Mean Teacher consistency learning with prototypical loss to enhance stage-aware embeddings, while a class-balanced focal loss addresses long-tailed distributions. Under limited labeling budgets (5–15% per class), the approach achieves robust performance (test accuracy 93.46%, macro-F1 91.86%, κ = 0.945, ROC-AUC 0.999), with Grad-CAM highlighting clinically meaningful intraparenchymal features. These results demonstrate the potential of a prototype-regularized semi-supervised pipeline to deliver accurate, interpretable fibrosis staging from routine US, while markedly reducing annotation demands. Future directions include cross-center validation, calibration, and prospective clinical evaluation. Thesis Thu, 11 Sep 2025 00:00:00 GMT http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/16740 2025-09-11T00:00:00Z http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/16739 Identification Of Medicinal Plants Using Deep Transfer Learning Saha, Shithi Rani The thesis explores the identification of medicinal plants in Bangladesh from images of plant leaves with the approach of deep transfer learning. The objectives of the study were to determine the best and less time consuming method to which individual could differentiate among the leafy medicinal plants and for what purposes. For this, I proposed and evaluated several deep learning architectures (MobileNet V2, Inception V3, ResNet50, VGG16 & VGG19). Performance from the test results suggested that the performance of the model outperforms all of the other models in MobileNet V2. On the dataset, it achieved an accuracy of 93.78%. In terms of performance, the precision 94%, recall and F1-score are 93%. Combining these results illustrates the efficiency of the model to predict the medicinal plants via leaf images. Thesis Thu, 11 Sep 2025 00:00:00 GMT http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/16739 2025-09-11T00:00:00Z http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/16738 Bangladeshi Currency Recognition and Counterfeit Detection Using Convolutional Neural Networks Baral, Saurav This thesis explores the recognition of Bangladeshi currency and the detection of counterfeit notes using deep learning. The primary objective of the study has been to discover if there is a reliable and time efficient solution where both genuine and fake notes of different denominations are identified. For this, I have also provided and compared different deep learning models like MobileNet V2, Inception V3, ResNet50, VGG16 and VGG19. Captioning Test Result Table results from test shows MobileNet V2 among the models are the highest. Its accuracy on the dataset was 96.11%. In terms of performance, the accuracy, recall and F 1 -score proved the robustness of such a prediction model. So, in summary, in this particular case we see the model does a good job in classifying real notes and fake notes. Thesis Sat, 11 Oct 2025 00:00:00 GMT http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/16738 2025-10-11T00:00:00Z