Research on identification of domestic commercial block cipher algorithms based on deep learning

With the widespread application of domestic commercial cryptographic algorithms and the advancement of commercial cryptographic application evaluation, the compliance of these algorithms has garnered significant attention. Various security agencies and research institutions in China have initiated studies on the identification of commercial block cipher algorithms and have explored their application in cryptographic evaluation work. This paper focuses on extracting features from ciphertext using the NIST randomness test method, followed by training and testing these features through various machine learning and deep learning methods. The paper consolidates relevant domestic research on this topic. In the final part of the study, encrypted data from the COCO2014 dataset using the domestic commercial cryptographic algorithm SM4 and the AES128 (CBC, ECB) algorithms are used for algorithm identification, employing MLP, CNN, LSTM, and Attention mechanisms. The experimental results demonstrate that CNN exhibits higher accuracy and stability compared to existing solutions, while the Attention mechanism shows advantages in subsequent AES128-ECB identification, albeit with highly sensitive to variations in the key-dimension selection.