Research and Analysis of Exam Cheating Detection Based on Deep Learning

With fast development of education, the problems of academic integrity in exams have become increasingly significant. Traditional manually invigilate has limited effectiveness in large-scale exams or online education environments, making it difficult to become real-time and high accuracy monitoring. Deep learning technology provides the possibility for automatic and intelligent cheating detection. This article reviews the research progress of the exam cheating detection based on deep learning, focusing on the application and development trends of object detection models (YOLO series), behavior recognition technology, and data augmentation methods (especially generative adversarial networks), and discusses the potential of multimodal fusion, lightweight deployment, and privacy protection strategies. This article provides an overall analysis from four aspects: dataset, model performance comparison, technical challenges and limitations, and future research directions. It then discusses the balance between model adaptability, real-time performance, and accuracy in the different scenarios, and points out the bottlenecks that still exist in current methods in terms of small samples, occlusion, complex behavior recognition, and ethical privacy protection. With the intention of provide reference and inspiration for research in this field, while also providing theoretical basics and guidance for the design, optimization, and promoting the intelligent invigilation technology in real educational scenarios.