http://ir.bdu.edu.et/handle/123456789/10165 Sat, 13 Jan 2001 07:32:54 GMT 2001-01-13T07:32:54Z http://ir.bdu.edu.et/handle/123456789/16529 MACHINE LEARNING BASED LABOR PREDICTION BY INCORPORTING CLINICAL INFORMATION WUDIE, BEKELE ALEMAYEHU Pregnancy is the period in which the female uterus carries a developing fetus. Labor is the detection of uterine contraction activity, which is monitored using various instruments such as ultrasound, IUP Cand, and tocodynamometry. However, the device or instrument is not an excellent estimation of the term or preterm birth date. We now have a low-accuracy device for monitoring maternal cases, which includes labor diagnosis and delivery prediction. It is difficult to make medical treatment decisions, including tocolytic therapy, the administration of steroids, and admission or transport to a hospital. There are many factors or conditions that are associated with the risk of preterm labor which includes stress, bleeding during pregnancy, and chronic conditions. In this study, shows that the maternal condition, which is the bleeding of the woman during pregnancy (early delivery), is the main factor or indicator of preterm labor, as well as, we have shown another alternative that is currently used in human monitoring techniques is uterine electro hysterography (EHG) or electromyography (EMG), this system Use an Ag/AgCl electrode between the above and below navel to record data. We stand by the clinical information of the EHG data set to demonstrate that this maternal condition is used for preterm and term labor prediction and analysis only in the first and second trimesters for preterm EHG signal features. To implement this work, it uses the TPEHG dataset from Physio Net, which contains only the first and second trimester maternal bleeding electro-hysterogram records obtained during regular examinations of pregnant women in the Department of Obstetrics and Gynecology of the Medical Center at the University of Ljubljana. by using machine learning algorithm for prediction labor, for preprocessing that utilizes a filter (Butterworth, IIR notch, Savitziky-Golay), and for feature extraction, (wavelet transform) and ten feature is extracted which is SSI, RMS, crest factor, variance, median frequency, mean power, peak to peak amplitude, peak frequency and sample of entropy. After this, we have use classifiers for classification, which are support vector machines (SVM), k-nearest neighbor (KNN), Decision tree (DT) and Random Forest (RF). We have achieved a result of SVM 84.85%, DT 81.82%, RF 84.85% and KNN 81.82% accuracy. Keywords: Electro hysterogram, Feature extraction, Labor, Machine Learning, Preterm birth, Support Vector Machine, Term birth, Wavelet Transform Thu, 01 Feb 2024 00:00:00 GMT http://ir.bdu.edu.et/handle/123456789/16529 2024-02-01T00:00:00Z http://ir.bdu.edu.et/handle/123456789/16528 Robust Trajectory Tracking control of Industrial Robotic manipulator Using Fuzzy Sliding Mode Controller Workineh, Ayal Robots are now commonly utilized to perform tasks more efficiently, accurately, and consistently than humans. Robotic manipulators are complex nonlinear systems characterized by strong interdependencies between multiple inputs and outputs, along with uncertainties and dynamic capabilities that change significantly over time. These factors contribute to the difficulties associated with controlling the trajectory of a robotic manipulator. To achieve optimal performance, the controller must address these issues. Sliding Mode Control (SMC) is a widely used method for such systems as it can effectively handle uncertainties and disturbances. However, designing an SMC system effectively requires precise knowledge of the extent of these uncertainties and disturbances, which can be difficult to obtain for complex robots with numerous degrees of freedom. The thesis develops a mathematical model for the robot manipulator's movement and torque using the well-known Euler-Lagrange equation to describe its dynamics. And also presents a controller design for a 3-DOF robotic manipulator that aims to achieve both accurate movement and robustness in the face of challenges. To ensure smooth operation and accurate positioning, the system's stability is analyzed using Lyapunov's stability criterion, which mathematically verifies if the robot's errors will dissipate over time. The thesis investigates the ability of a robot arm to follow various paths using a fuzzy sliding mode controller (FSMC), focusing on the controller's robustness in handling unexpected challenges. The simulation Result shows that the proposed controller is strong and effectively tracks trajectories, maintaining manipulator joint angle errors at approximately 10-3(rad) at steady state. Additionally, it has reduced Integral Time Absolute Error (ITAE). Furthermore, the fuzzy sliding mode controller (FSMC) has been observed to have lower control input torques compared to the Sliding Mode Controller (SMC), and it eliminates the chattering effect. Keywords: Three degree of freedom robotic manipulator, Trajectory tracking, sliding surface, Sliding mode control, fuzzy sliding mode control, Sat, 22 Jun 2024 00:00:00 GMT http://ir.bdu.edu.et/handle/123456789/16528 2024-06-22T00:00:00Z http://ir.bdu.edu.et/handle/123456789/16527 DEEP LEARNING BASED CRYPTOGRAPHY KEY GENERATION FROM BI-MODALBIOMETRIC FACE AND FINGERPRINT TAHAYU, GIZACHEW YIRGA In modern digital security systems, stochastic random processes and internal mathematical transformations are widely used in the creation and maintenance of encryption keys, such as passwords and PINs. Despite offering robust protection, these keys need complicated and costly systems for distribution and storage. In order to avoid the requirement for complex storage and distribution procedures, this study investigated a different strategy that generates encryption keys using biometric data. For guaranteeing the security and dependability of the generated keys, the key generation process is made to be resistant to noise, changes, and assaults on the sensor data. A set of combined biometric face images and fingerprint data was used for experiments that show how well and reliably the proposed system works at making strong cryptographic keys. The study explores biometric key generation techniques based on deep learning models, specifically Facenet and VGG19 with PCA for dimensional reduction convolutional neural networks used to extract biometric features from human facial images and fingerprint images, respectively. We combined the extracted features and divided them into two groups: train and test. The developed Siamese Neural Network (SNN) model based on this dataset showed promising results, with train and validation loss reducing from 0.35 to 0.04 and 0.3 to 0.03, respectively. Measured using vector converter sigma similarity and sigma difference, the accuracy reached 99.8% and 46.0%, respectively. The results indicate that the system achieves high key generation rates while maintaining low error rates, with False Acceptance Rate (FAR) and False Rejection Rate (FRR) less than 1% and 2.7%, respectively. This makes it suitable for use in secure authentication systems that require strong and reliable keys. Overall, this thesis contributes to the advancement of biometric security systems by using a deep learning-based approach with code-based cryptography for generating secure keys from fused biometric data. Key words: Deep Learning, SNN model, Facenet, VGG19, PCA, Coded based Cryptography, Biometric, and Fusion Wed, 01 Jan 2025 00:00:00 GMT http://ir.bdu.edu.et/handle/123456789/16527 2025-01-01T00:00:00Z http://ir.bdu.edu.et/handle/123456789/16526 PERFORMANCE COMPARISON OF CONCATENATED (SLM + CMA + CLIPPING) PAPR REDUCTION TECHNIQUE WITH INDIVIDUAL ALGORITHMS FOR MIMO-OFDM SYSTEM NOLAWIT, STOTAW KASSA Multiple input multiple output orthogonal frequency division multiplexing (MIMO_OFDM) systems have become a key component of wireless communications in the attempt for effective data delivery. The high Peak to Average Power Reduction (PAPR) of these systems, however, poses a number of fundamental difficulties and can seriously reduce power efficiency and result in non-linear distortion. In order to address the PAPR issue, this paper presents a novel method that combines the Constant modulus algorithm (CMA), Clipping, and Selective Mapping (SLM). The sequence with the lowest PAPR is chosen for transmission after the SLM algorithm creates many candidate signal sequences with different phase shifts. CMA is utilized to keep the modulus of the signal constant, and clipping is applied to cut signals whose peak amplitudes beyond a predetermined threshold. the concatenated method has the best PAPR reduction performance with different antenna configuration and different subcarriers. Keyword: Clipping, Constant modulus algorithm (CMA), Multiple input multiple output orthogonal frequency division multiplexing (MIMOOFDM), Peak to Average Power Reduction, Radio Frequency Amplifier (RF) and Selective Mapping (SLM) Sat, 01 Jun 2024 00:00:00 GMT http://ir.bdu.edu.et/handle/123456789/16526 2024-06-01T00:00:00Z