融合DropBlock和注意力机制的视频异常检测算法OA

Video anomaly detection is one of the key tasks in the field of computer vision.Traditional video anomaly detection methods are susceptible to background noise interference,difficult to effectively capture local details,and prone to poor generalization ability and insufficient robustness due to overfitting training data.In order to solve these challenges,we propose a video anomaly detection algorithm that combines DropBlock and attention mechanism.Based on the U-Net architecture,the Squeeze-and-Excitation Module and the Spatial Attention Module are introduced into the bottleneck layer and the jump connection,respectively.The SE module enhances the feature representation of important channels through the channel attention mechanism,while the spatial attention module increases the focus on key regions by dynamically adjusting spatial weights.The Transformer is integrated after the SE module to enhance the model's ability to model the spatio-temporal features of videos.At the same time,by introducing DropBlock into the convolutional layer,the overfitting problem of the convolutional network is effectively alleviated and the generalization ability of the model is enhanced.The ex-perimental results show that the AUC indexes of the proposed method on UCSD-Ped2,CUHK Avenue and ShanghaiTech public datasets reach 96.9％,86.2％and 73.1％,respectively,which verifies its effectiveness.