This paper tackles the limitations of traditional network security assessment methods, which suffer from weak feature representation and low classification accuracy. The proposed approach uses a convolutional denoising autoencoder (CDAE) to enhance feature extraction from attack data, with a channel attention mechanism added in the decoder to retain critical spatial information. Additionally, a BiGRU with residual connections is utilized to better extract and preserve contextual information. The network security situation is assessed by calculating a value based on attack severity and impact. Experimental results show that this method significantly outperforms existing models in accuracy, precision, recall, F1-score, and mean square error, proving its effectiveness for large-scale, high-dimensional data. This study is the first to combine CDAE, channel attention, and residual BiGRU, providing new insights into feature extraction and classification for network security. Future work may evaluate its robustness on varied datasets.