<p>Video anomaly detection is an important and challenging task in the field of computer vision. Current video anomaly detection methods are mostly based on reconstruction, prediction, or a combination of both. This paper proposes a reconstruction-based detection method aimed at improving anomaly detection accuracy by enhancing the structure of the encoder and decoder. The method leverages a memory-enhanced reconstruction approach, exploring new video anomaly detection pathways through the use of channel attention and spatial attention. First, a new structure for the encoder layer is designed, employing an expanded memory module and added skip connections, allowing the model to more effectively retain the information from the original input data. Next, we introduced the Local–Global Module into the encoder, which combines Convolutional Neural Networks (CNN) with the self-attention mechanism of Transformers, enabling the model to capture both local features and global contextual information, thereby enhancing its understanding of spatiotemporal dependencies. In the decoder, we parallelly integrate the PA module and EA module. The PA module strengthens the processing of complex spatiotemporal information through interactions between local and global features, while the EA module enhances the model’s responsiveness to key feature channels using an adaptive channel attention mechanism.The structural improvements in the encoder and decoder enable the model to more accurately detect abnormal behaviors during anomaly detection, significantly improving the detection performance. Experimental results show that this method outperforms existing methods on various standard datasets.</p>