<p>Large model has become a hot topic in academia and industry area, whose infrastructure includes lots of data transmission and transformation technologies. To prevent malicious attacks, data encryption is used commonly for data protection. However, with the AI technology development, the feature of encrypted data can be extracted and analyzed more easily than before. To study the impact of different backdoor attacks on classification and find a defense in the future. Therefore, in this paper we extend a steganographic backdoor attack approach based on the Least Significant Bit (LSB) to attack encrypted traffic; more specifically, the traffic data files are transformed into two-dimensional grayscale images after data processing and normalized by the USTC-2016 toolset and two-dimensional array processing by the Numpy toolset in Python; then we use LSB Steganography to attack the traffic images and poison the training dataset and analyze the neural network for classification tests after contamination and many experiments are performed with different portions of poisoning data, which indicates that it has an effect on the stealthiness and effectiveness of the attack.</p>