A Novel Detection Model Based on Improved YOLOv11n with Spatial Attention Downsampling

Using deep learning technology to detect exam cheating behaviors can greatly improves the efficiency and quality of proctoring, as well as promoting fairness and justice in examinations. However, current research primarily concentrates on developing detection systems that rely on existing algorithms, which exhibit limited detection capabilities for small targets and subtle, concealed cheating behaviors in real- world scenarios. To address this challenge, this paper analyzes the downsampling methods of the YOLO series and identifies that existing methods have overlooked the loss of spatial information during down-sampling. Consequently, we propose a downsampling method that enhances spatial attention learning, namely SpAttDown. Furthermore, based on this, the SpAttDown_YOLOv11n algorithm has been developed. Finally, comparative experiments on three exam behavior datasets show that the proposed algorithm achieves the highest detection accuracy while maintaining a fast detection speed.