Network intrusion detection technology, as a critical security defense mechanism, plays a pivotal role in the timely detection of network attacks. However, existing intrusion detection algorithms exhibit limitations in multiclass classification accuracy, the ability to capture the complex relationships within input sequences, and the effective handling of long sequences. In this paper, we propose a novel deep learning model, the B2-CTA model, which integrates Convolutional Neural Networks (CNN) and the Transformer architecture to extract spatial features and capture long-range dependencies within the input data. Additionally, it employs a Bidirectional Long-Short-Term Memory (BLSTM) network to extract temporal features. To enhance model efficiency, random forests are utilized to identify and select the most significant features as inputs, thus mitigating redundancy in input data. Furthermore, a self-attention mechanism is introduced to assign varying weights to the input data based on their importance, enhancing detection accuracy. Finally, the classification results are obtained through a Softmax classifier. We perform multiclass classification tests and evaluations on the widely used NSL-KDD and UNSW-NB15 datasets, and the experimental results demonstrate that the proposed model achieves accuracies of 99.31% and 83.04% on these two datasets, respectively.