The cross domain generalization problem of image classification is a common challenge in computer vision. For example, when performing image classification under different weather, lighting, and scene changes, the model performs well in the training domain, but performs poorly in unseen new domain environments. The root cause of this phenomenon lies in the differences in data distribution in different fields, which makes it difficult for the model to adapt to new distributions, resulting in insufficient generalization ability. To address this issue, domain adaptation technology has become an important research direction. The goal is to enable the model to automatically adapt to new data distributions and maintain high classification accuracy without the need for retraining. This article proposes a cross domain adaptive method based on category and feature alignment. This method is based on two core technologies, adversarial domain adaptation and category alignment, and combines self supervised learning and feature reconstruction strategies to enhance the domain generalization ability of the model. In the training phase, adversarial training is used to minimize the difference in feature distribution between the source domain and the target domain.