In today’s information age, network malicious attacks have become a serious security issue, posing a significant threat to data integrity, confidentiality, and availability. In order to effectively detect and respond to these attacks, this paper pro-poses a network malicious attack detection method HSAM (Hierarchical Spatial Attention Mechanism) based on a hierarchical spatial attention mechanism. By constructing a deep learning model and integrating a hierarchical spatial attention mechanism, this method can capture the spatiotemporal features of network traffic data, significantly improving the accuracy of malicious attack detection. The HSAM model uses a hierarchical spatial attention mechanism to weight the important features in network traffic data, enhancing the model’s focus on key features and reducing interference from redundant information. Finally, feature output is achieved through the GRU layer, enabling efficient detection of network malicious attacks. Experimental results demonstrate that this method outperforms traditional detection methods on public datasets, with high accuracy, precision, and recall rates. This research provides an effective means of detecting malicious attacks in the field of network security, and also showcases the potential and value of attention mechanisms in network security applications.