The intelligent development of automobiles has increased the dependence on automotive bus networks, such as real-time power control and maneuvering control of automobiles, all of which require the use of in-vehicle networks as a medium for information transfer. Due to the lack of security measures such as information authentication and identity authentication in Controller Area Network (CAN), it is easy to be invaded by hackers. Therefore, in order to improve the security communication guarantee of in-vehicle CAN network, this paper proposes a dual detection method based on the combination of information entropy and CNN-SVM. The method judges the suspicious traffic by preliminary detection based on information entropy, and then further makes more accurate attack judgment on the suspicious traffic based on the deep detection method of CNN extracting features and SVM for classification. The experimental results show that the dual detection model improves the detection effect and has better performance compared with the single model.