Improved Cheating Detection in Examinations using YOLOv8 with Attention Mechanism

• المؤلف الرئيسي: YAN, ZUO
• التنسيق: أطروحة
• اللغة: الإنجليزية; الإنجليزية
• منشور في: Universiti Malaysia Sarawak 2025
• الموضوعات: QA75 Electronic computers. Computer science
• الوصول للمادة أونلاين: http://ir.unimas.my/id/eprint/48331/

Examinations are pivotal in educational and talent-selection processes globally. Ensuring their integrity is critical, but traditional invigilation methods, combining manual oversight with video monitoring, are resource-intensive and not fully effective in detecting cheating. This thesis presents an innovative approach to enhance examination fairness through automated cheating detection using an improved YOLOv8 algorithm integrated with an attention mechanism. The study focuses on developing and implementing a target- detection system capable of identifying student abnormal behaviors indicative of cheating in real-time. This system utilizes the advanced capabilities of the YOLOv8 model, optimized for speed and efficiency, making it suitable for deployment on regular- performance computers. The integration of an attention mechanism allows the system to focus on key visual cues that signify potential dishonesty, thereby improving detection accuracy. An original dataset, representing various cheating methods in paper-based exams, was manually compiled due to the absence of suitable open-source data. This dataset included a wide range of cheating behaviors, enabling comprehensive training and validation of the model. Results demonstrate that the improved YOLOv8 model achieves a detection accuracy of 82.71%, significantly reducing the need for manual video review and labor costs associated with traditional invigilation methods. This high accuracy rate meets the practical application requirements for real-time cheating detection in offline exam venues. This research contributes to the field of educational technology by offering a scalable, accurate, and efficient solution to a prevalent challenge in academic assessment. The successful implementation of this system can revolutionize examination monitoring, ensuring fairness and upholding academic integrity. Future work could extend this model's application to digital or oral examinations and explore its integration with other surveillance technologies. The proposed system's ability to function on standard computer hardware highlights its practical utility, while the incorporation of an attention mechanism ensures precise identification of cheating behavior amidst normal exam conditions. By addressing the limitations of existing invigilation practices, this study paves the way for more reliable and equitable examination environments. Ultimately, the adoption of such advanced technologies could establish new standards in academic integrity, benefiting educational institutions and students alike.