Faculty of Technoscience

Permanent URI for this community

https://dir.muni.ac.ug/handle/20.500.12260/2

Browse

Browsing Faculty of Technoscience by Subject "Accuracy"

Now showing 1 - 5 of 5
  • Thumbnail Image
    Item
    AI-driven pathogen detection systems for rapid and accurate diagnosis
    (IEEE, 2025-09-24) Deepika, M.; Murugan, Sundara Bala; Subasini, V.; Rufus, N. Herald Anantha; Atkinswestley, A.; Ali, Guma
    Pathogen detection at speed along with precision serves as a basis for disease diagnosis and overall control effectiveness. Several current detection methods based on culture-based techniques and PCR face drawbacks of extended processing times, high expenses, and the need for expert operators. It adopts deep learning Convolutional Neural Networks with an attention mechanism in a system designed to overcome existing limitations of pathogen detection. The model was trained using a multi-modal dataset that combined spectral and biological signals through feature extraction optimization and privacy-protected training methods based on federated learning. The proposed system reaches a 96.8% detection accuracy together with 15% reduced false alarm rates and 35% faster operation speeds than conventional models. Experimental data measurements showed that this system achieved 92.3% F1-score together with 97.1% AUC for real-time identification of pathogens. AI has proven its effectiveness through these findings in changing the way pathogen diagnostics operate. Further work will concentrate on extending datasets while creating real-time deployment protocols for clinical use.
  • Thumbnail Image
    Item
    FPGA-based High-precision MCU testing using pattern latching and DDLL
    (IEEE, 2025-07-25) Arunkumar, K; Vani, R; Ali, Guma
    Performance testing of microcontrollers (MCUs) is very crucial especially when it comes to safety critical systems such as those in automotive electronics field that will require high accuracy and reliability. One of the essential criteria for performance testing are results checking for getting the maximum possible clock frequency of the MCU possible to influence the automobile units’ efficiency. The traditional method of testing using the ring oscillator approach might not be sufficient in giving MCU performance. To this, the system under consideration includes a pattern latching algorithm with an included configurable ring oscillator. The utilization of a clocked latch with a synchronous gate is used to synchronize ring oscillator circuits so as to do precise pattern analysis. The system makes use of a digital delay locked loop (DDLL) to provide high accuracy when measuring performance by dynamically compensating for delays. Such a strategy increases the validity of performance screening, with improved estimation of the MCU’s operating abilities in adverse conditions.
  • Thumbnail Image
    Item
    Machine learning for medical image feature extraction
    (IEEE, 2025-05-09) Rathore, Saurabh Pratap Singh; Ali, Guma; Chamoli, Sakshi; Lotus, Rayappan; Kumar, Yogendra; Sikarwar, Shailendra Singh
    Feature extraction from medical images is crucial for harnessing the vast information they contain, aiding in diagnosis, treatment planning, and disease monitoring. Traditional feature extraction methods often struggle to capture the complex patterns and subtle variations in medical images. Recently, machine learning techniques have become powerful tools for automatically extracting discriminative features, enabling more accurate and efficient analysis. This paper provides a comprehensive review of advanced machine learning approaches for medical image feature extraction. It covers various methods, including deep learning architectures, convolutional neural networks (CNNs), and feature learning techniques, highlighting their applications across different medical imaging modalities such as MRI, CT, and X-ray. Our CNN model achieves an average classification accuracy of approximately 94%, outperforming the pneumonia detection accuracy of KNN (91%) and SVM (92%).
  • Thumbnail Image
    Item
    Multi-modal biometric authentication integrating gait and face recognition for mobile security
    (IEEE, 2025-10-28) Nivedita, V.; Joseph, Juno Ann; Varghese, Divya Thankom; Nithyanandh, S; Sundaram, Jayavelu; Ali, Guma
    In the surroundings of mobile security systems, traditional biometric methods such as fingerprint and facial recognition can be readily disputed and misled. These systems change user appearance by failing in different illumination conditions or spoofing attacks with false fingerprints or images. Overcoming these constraints, the proposed multi-modal biometric identification system integrates gait with facial recognition. Combining facial data with gait traits observed by accelerometers and gyroscopes helps to increase security and reduce false positives and negatives. Results show the system achieves 97.8% accuracy, significantly above the existing systems using a low false acceptance rate (significantly) of 1.5% and a false rejection rate (FRR) of 0.7%. It has a superior resilience to spoofing techniques, including stride imitation and image/video spoofing; it also shines in tough conditions, including poor lighting and appearance alterations. Despite a slight increase in processing time, the method offers a more robust and reliable solution for mobile authentication.
  • Thumbnail Image
    Item
    Next-gen non-invasive glucose monitoring using microwave sensors and ai-based thumb positioning analysis
    (IEEE, 2025-06-25) Uma, S.; Soundharya, P.; Charisma, S.; Jackson, Beulah; Ganapathi, G.; Ali, Guma
    Because of discomfort and accuracy problems, the existing non-invasive glucose monitoring devices include optical sensors and finger-prick tests have long been under doubt for their viability. Variations in skin type and environmental interference might make these techniques unreliable. To get past these constraints, the study presents a fresh approach combining artificial intelligence (AI)-based thumb positioning analysis with microwave sensors. By spotting changes in the dielectric characteristics of the skin brought on by glucose, the microwave sensors offer exact, invasive, real-time readings. Through best sensor alignment, AI removes thumb position errors. Performance assessment indicates substantially better outcomes than the existing systems with an R2 of 0.98, RMSE of 7.1 mg/dL, and MAE of 5.2 mg/dL. The great flexibility of the proposed system to a wide spectrum of demographics and excellent user compliance (95%) underlines its possibility for efficient and comfortable diabetes control. The technique represents a substantial development in non-invasive glucose monitoring.