http://ir.bdu.edu.et/handle/123456789/10140 The diffusion of technology and knowledge is a salient feature in the technological change, innovation and growth of the modern society. Much attention is being given to the role of universities in the development of nations. Universities all over the world are playing a crucial role in producing competent professionals who will be taking responsibility of the country’s technology and resources. This goal of universities can be further fostered by a strong link with the industries. A strong bond and relation between universities and the industry helps not only in finding out immediate solutions to prevailing challenges in the industry but also will help in producing all rounded and sharp professionals who will later be important assets to the development of the country.The Faculty experienced a surge in enrolment of students in the regular as well the continuing program, which reveals the fact that the Faculty offers students not only the excellence in cutting-edge technological knowledge and applications, but also in a multidisciplinary engineering and science background for the career that our students have planned. Share Sat, 13 Jan 2001 07:33:13 GMT 2001-01-13T07:33:13Z http://ir.bdu.edu.et/handle/123456789/16483 Fine-grained Bone Fracture Classification to Enhance Surgical Diagnosis and Medication Dereje, Mulugeta Human bones serve as protective structures for vital organs, and fractures represent either complete or partial breaks in these bones. This study aims to classify bone fractures to enhance surgical diagnosis and treatment. It specifically addresses five types of fractures: comminuted, fracture dislocations, oblique, pathological, and spiral fractures. While also considering multi-region fractures. The classification utilizes X-ray images, which are commonly used in medical settings for diagnosing fractures. To accomplish this, we used a hybrid approach that integrates various image processing techniques and machine learning methods. This includes using k-means and watershed algorithms for segmentation, which effectively isolates the fracture areas from complex background images. For feature extraction, we applied Convolutional Neural Networks (CNNs), enabling the automatic identification of relevant features from the segmented images. We then used Support Vector Machine (SVM) classification to accurately categorize the different types of fractures based on these extracted features. We thoroughly tested the proposed method and achieved an accuracy of 95%. This high accuracy shows that combining advanced image processing techniques with machine learning can greatly improve diagnosis in orthopedic settings, by helping doctors make faster and more accurate treatment decisions for patients with bone fractures; this study improves the reliability of fracture classification in clinical practice. Keywords: Bone Fractures, Image Processing, CNN and SVM. Tue, 01 Oct 2024 00:00:00 GMT http://ir.bdu.edu.et/handle/123456789/16483 2024-10-01T00:00:00Z http://ir.bdu.edu.et/handle/123456789/16482 CONSTRUCTING A MODEL FOR STROKE DISEASE TYPE IDENTIFICATION USING TRANSFER LEARNING BIRARA, MAREW Stroke is a leading cause of morbidity and mortality worldwide, necessitating prompt and precise diagnosis to facilitate effective treatment. This thesis presents a comprehensive study on the identification and classification of stroke types using brain CT scan images through deep learning techniques. Utilizing a dataset of brain CT scans labeled as ischemic or hemorrhagic strokes, we employed advanced preprocessing steps, including normalization, noise reduction, and image enhancement, to optimize image quality. A strategic noise removal process was implemented, starting with the assessment of noise presence using Peak signal-to-noise ratio (PSNR) values. Identified noise types included Gaussian, Speckle, Poisson, Exponential, and Gamma noise. Effective noise filtering techniques such as Median and Non-Local Means filters were applied, significantly increasing PSNR values. Histogram equalization was used for further image enhancement. Additionally, rotation correction was performed using the midsagittal principle to ensure the correct orientation of images. We trained and fine-tuned several deep-learning models, including EfficientNetB0. The models' performances were remarkable, with EfficientNetB0 achieving an accuracy of 100% and perfect precision, recall, and F1-score, demonstrating its superior classification capability. Overall, this study underscores the effectiveness of advanced preprocessing techniques and parameter tuning in enhancing the performance of deep learning models for stroke classification, ultimately improving diagnostic accuracy and patient outcomes. This study's limited sample size and lack of MRI data constrain its ability to fully analyze stroke characteristics and outcomes, affecting the generalizability of the findings. Future research should include a larger, more diverse cohort and incorporate MRI data to enhance the robustness and applicability of the results. Keywords: Stroke Disease, Medical imaging, Brain CT scans, Transfer learning, Stroke Disease Identification. Sat, 01 Jun 2024 00:00:00 GMT http://ir.bdu.edu.et/handle/123456789/16482 2024-06-01T00:00:00Z http://ir.bdu.edu.et/handle/123456789/16481 Amharic Visual Question Answering on Ethiopian Tourism Alebachew, Molla Dinku Visual Question Answering (VQA) is a Vision-to-Text (V2T) task that integrates visual features of images with natural language questions to generate meaningful responses. Most existing research has focused on English, leaving a significant gap for other languages, including Amharic. Tourism, a major global industry, relies heavily on interactions where visitors seek information about natural, historical, cultural, and religious sites. Ethiopia is a remarkable tourist destination, home to unique sites most visitors are local, creating an urgent need for a VQA model that can deliver accurate, culturally relevant information in Amharic. Unfortunately, no such model currently exists to assist tourists at these heritage sites. This research addresses this gap by developing an Amharic Visual Question Answering model specifically tailored for Ethiopian tourism. A new Amharic VQA dataset was created using 2,200 diverse images from Ethiopian tourist sites paired with 6,600 questions in Amharic. Our dataset is collected from various sources, including the UNESCO website, the Amhara Tourism office, and online platforms such as Facebook, Free pixel, and Instagram. Each image is complemented by three corresponding questions formulated by three individual experts and answered by ten candidates. The questions, answers, and images are linked through annotations and fed into the model. We used ResNet-50 for feature extraction and Bidirectional Gated Recurrent Unit (BiGRU) with attention mechanisms, achieving a testing accuracy of 54.98%, demonstrating the model's effectiveness in answering questions about Ethiopian heritage. We will expand this research using external knowledge to get answer and description beyond image and custom object detection. Key word: Amharic Language; Ethiopian tourism; Deep learni Fri, 01 Nov 2024 00:00:00 GMT http://ir.bdu.edu.et/handle/123456789/16481 2024-11-01T00:00:00Z http://ir.bdu.edu.et/handle/123456789/16480 AMHARIC SPEECH TO ETHIOPIAN SIGN LANGUAGE TRANSLATION SELAMAWIT, BELAY Ethiopian Sign Language (EthSL) is the primary means of communication for the deaf community in Ethiopia. The dearth of accessible tools that convert Amharic speech into EthSL, however, makes it difficult for hearing and deaf people to communicate. The goal of this study is to create a deep learning model for translating Amharic speech to video sign language. To close this gap, this study attempts to create a deep learning-based Amharic speech-to-video sign language translation model. Four deep learning models were compared: Long Short-Term Memory (LSTM) networks, Variational Autoencoders (VAEs), Recurrent Neural Networks (RNNs), and Convolutional Neural Networks (CNNs). Via hyperparameter optimization, LSTM achieved the highest accuracy, increasing from 91% to 97%. In order to ensure model performance, interpretability, and usability, the study used a design science methodology. To increase the model's decisionmaking process's transparency, integrated gradients were used. A Flutter-built mobile application including the trained LSTM model was evaluated for usability using surveys. Results show that as compared to previous deep learning models, the LSTM-based model considerably increases the accuracy of Amharic speech-to-EthSL translation. The prototype improves accessibility for the deaf community by enabling real-time EthSL video creation from Amharic speech. By offering a dependable and understandable solution, encouraging inclusivity, and setting the stage for future developments in Amharic speech-to-sign language translation, this research advances assistive technology. For this thesis, four deep learning algorithms were investigated: Long Short-Term Memory (LSTM) networks, Variational Autoencoders (VAEs), Recurrent Neural Networks (RNNs), and Convolutional Neural Networks (CNNs). After hyperparameter optimization, LSTM outperformed the others, increasing its accuracy from 91% to 97%. Using a design science methodology, this study guarantees interpretability and usability in addition to model performance. Transparency in Amharic speech-to-video sign language translation was improved by using Integrated Gradients to explain the model's decision Sun, 01 Jun 2025 00:00:00 GMT http://ir.bdu.edu.et/handle/123456789/16480 2025-06-01T00:00:00Z