With the rapid development of society, network technology continues to advance. At the same time, people are also facing threats from malicious traffic. To address this issue, this paper proposes a malicious traffic detection method based on word vectorization algorithm and neural networks. This method combines tasks such as packet parsing, header data grayscale image generation, and message feature extraction to extract effective feature information. By combining convolutional neural networks (CNN), it detects malicious traffic hidden within normal traffic. Experimental results demonstrate that compared to traditional machine learning methods, the proposed approach exhibits higher efficiency and accuracy in identifying malicious traffic attacks, thereby better-protecting network users from malicious attacks.