Vehicle communication is recognized as a significant application field for high-speed serializer and deserializer(SerDes) technology. To address the issue of severe inter-symbol interference caused by high-frequency attenuation in complex vehicle environments, particularly under higher-order modulation schemes, a deep learning approach was introduced. A low-complexity channel equalization scheme based on the Transformer encoder structure was designed to enhance the quality of received signals. In this scheme, the input sequence was transformed into abstract representation vectors, followed by the extraction of feature information from these vectors through encoder layers. Subsequently, a fully connected layer classified the signals based on the extracted feature information, thereby achieving high-speed SerDes channel equalization. Experimental results demonstrate that compared to traditional adaptive algorithms and fully connected neural network models, the proposed scheme effectively reduces signal distortion caused by high-frequency attenuation. The scheme achieves a computational complexity reduction of 19% and 24%, respectively, while providing received signal-to-noise ratio gains of 1.8 dB and 0.9 dB. By applying the proposed channel equalization scheme in high-speed SerDes systems, the signal transmission quality has been improved, and the system's robustness has been enhanced.