Traditional intrusion detection methods have limitations in dealing with complex data and new threats. This paper proposed a deep learning intrusion detection model based on SVMSMOTE oversampling and convolutional neural network (CNN) to improve the defense capability and detection accuracy of the system. In the preprocessing stage, the SVMSMOTE method is used to oversample the imbalanced dataset to achieve data balance, and a feature importance evaluation method based on a random forest classifier is used for feature extraction. During model training, Focal Loss is introduced as a loss function to improve the generalization ability and accuracy of the model across the entire dataset. Test results show that the proposed method achieves an intrusion detection accuracy of 97.91%, which is excellent.