In video anomaly detection tasks,while memory-augmented autoencoder frameworks effectively suppress overfitting of anomalous samples,they often fall short in capturing fine-grained details from normal ones.To address this issue,this paper proposes a video anomaly detection model enhanced by multi-scale wavelet transforms.Built upon a U-Net backbone network,the model achieves superior performance through two key innovations:1) a Haar wavelet transform-based downsampling module in the decoder compresses redundant information while retaining critical details via multi-resolution decomposition in frequency domain;2) a Discrete Wavelet Transform Feature Fusion (DWTF) mechanism is introduced within the cross-layer skip connections,leveraging the complementary enhancement of structural representations through low-frequency and high-frequency components across scales.Experimental results demonstrate that the proposed method achieves AUC scores of 97.0% and 88.2% on the UCSD Ped2 and CUHK Avenue datasets,respectively.Ablation studies further verify that the wavelet downsampling and feature fusion modules contribute performance gains of 0.9 and 2.8 percentage points on the CUHK Avenue dataset,validating the effectiveness of multi-scale frequency-domain feature modeling.