http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/34 Sat, 09 May 2026 16:10:40 GMT 2026-05-09T16:10:40Z http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/17166 A Deep Learning-Based Waste Classification Using Ensemble and Vision Transformer Models Ahmmed, Sayem; Tasnim, Zarrin This research presents a deep learning approach for waste categorization using three pre-trained models such as MobileNetV2, DenseNet121, and ResNet50, a transformer model (Vision Transformer or ViT), and an ensemble model. The seven-class waste dataset was used, which is publicly available, with the preprocessing steps including resizing, normalization, augmentation, and class balancing. Hyperparameter tuning was applied to all models using Grid Search, Random Search, and Bayesian Optimization. Among them, the ensemble model had a test accuracy of 97.52%, surpassing single models by synergistically combining their predictions by weighted averaging soft voting. The models were made robust using label smoothing, mix-up augmentation, and class weighting. Evaluation was carried out on accuracy, precision, recall, F1-score, and confusion matrices. Issues such as class imbalance and intra-class visual similarity in visual waste classification are addressed by the study. The future work will use the system in an IoT-capable intelligent dustbin for actual implementation. Project Report Wed, 14 May 2025 00:00:00 GMT http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/17166 2025-05-14T00:00:00Z http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/17165 Efficient Jute Disease Classification Using Hybrid Deep Learning Model Rahman, Md. Waysur Jute, being one of the most important economic crops in South Asia, is greatly impacted by leaf diseases such as Cercosporin Leaf Spot and Golden Mosaic, which traditional diagnosis tools are not well-equipped to address. While deep learning has revolutionized plant disease detection, existing frameworks for jute are hindered by small datasets, computational inefficiency, and unfiled applicability. In this paper, we propose an efficient deep learning architecture for autonomous jute disease classification, integrating novel preprocessing, best transfer learning, and real-time deployment. We introduce a robust dataset of 12,000 high-resolution images in three categories (healthy, Cercosporin Leaf Spot, Golden Mosaic), class-aware augmented with data sparsity and imbalance alleviation. Our method employs a hybrid preprocessing pipeline of wavelet-based denoising (Daubechies-4) and adaptive color normalization to disentangle leaf regions and make use of discriminative features. Using a ResNetRS50 model with transfer learning fine-tuning, we obtain state-ofthe-art results with 98.5% validation accuracy (4.3% improvement over existing literature) and 97.8% precision for challenging field images under varying illumination and occlusion. The model performs real-time inference at 42 FPS on NVIDIA T4 GPUs with a light footprint (1.2 GB VRAM) and is compatible with deployment on edge devices. Technical innovations include a dynamic augmentation method balancing minority classes through synthetic lesion generation, in-pipeline explain ability through Grad-CAM visualizations for farmer-friendly diagnosis, and a multi-stage training protocol combining progressive resizing and label smoothing to enhance generalization. Experimental validation on 1,850 samples from three geographies confirms 96.4% operating accuracy, surpassing human expert prediction by 23% in identifying early-stage disease. The applicability of the system is demonstrated by a simulated 18–22% reduction in crop loss through on-time intervention. We openly release the dataset and model to encourage further work, establishing a new benchmark for crop-specific AI technology in precision agriculture. Project Report Wed, 14 May 2025 00:00:00 GMT http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/17165 2025-05-14T00:00:00Z http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/17164 Enhancing Rice Plant Disease Detection: A Classification Approach with Transfer Learning Mishu, Mrinmoy Saha Rice is the dominant crop of Bangladesh. Yet, every year a large number of crops are affected by the disease. Conventional detection of rice plant disease in the lab is timeconsuming and involves a lot of time. It leads to a decrease in production and economic loss for farmers. Therefore, image-based disease classification is a rising research discipline. In this paper, we've taken an approach to classify the nine frequent rice diseases and normal rice plant leaves by image. With transfer learning and model finetuning, we classified a total of ten classes. We are using Convolution Neural Network (CNN) and Deep Learning (DL), a subfield of Artificial Intelligence, for doing this kind of classification with automatism by training an over ten-thousand image list. We've achieved 98.47 percent validation accuracy with EfficientNetB3 of total ten class where nine of them are disease class and remaining one is normal. After identifying the bestperforming model, we converted it using TensorFlow Lite model maker for deployment in a mobile application. This app enables real-time disease detection by allowing users to capture or select an image, which the model then classifies instantly. This paper includes methodology and how we achieved validation accuracy along with previous literature on this. Discussion on hyperparameter tuning and utilization of different categories of pretrained models which were trained on 'ImageNet' are present. Project Report Wed, 14 May 2025 00:00:00 GMT http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/17164 2025-05-14T00:00:00Z http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/17163 Exploration of Anemia in Hematology Patients: Using Artificial Intelligence Present and Future Perspective Hasan, Mahamudul Anemia is a public health disease that reflects the lack of red blood cells or hemoglobin to carry the requisite amount of oxygen to the body, causing weakness, fatigue, and diminished cognition. It is a gigantic public health concern in developing countries like Bangladesh, where the timely diagnosis is still hampered because of less availability of resources and low awareness amongst common people. The study begins to create an intelligent and accurate anemia prediction model using various machine learning models with patient data from a Bangladesh general hospital. The methodology involved a systematic preprocessing of the dataset, including dealing with missing values, normalization, and categorical encoding, and splitting the dataset into training data and test data. Ten historical classifiers were utilized: Random Forest (RF), Logistic Regression (LR), Gaussian Naïve Bayes (GNB), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Quadratic Discriminant Analysis (QDA), Ridge Classifier (RC), Passive Aggressive (PA), XGBoost (XGB), and Grid Search CV (GS). Bagging, boosting, and voting were carried out to ensemble and improve classifiers. Accuracy, precision, recall, and F1 score were used as the measurement metrics. According to the findings, the Voting classifier performed better than all others in performance with the highest accuracy of 93.12%, followed by KNN, GS, and RF under bagging and the default setting. Boosting generally delivered a mixed performance with overfitting shortening some of the models. The study concludes. Project Report Wed, 14 May 2025 00:00:00 GMT http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/17163 2025-05-14T00:00:00Z