With the development of digital economy, Edge intelligent services have become an important component of the future 6G network. Due to the high dynamic mobility of intelligent inspection equipment during operation and inspection in IIoT (Industrial Internet of Thing), Edge intelligent models are confronted with the problems of data drift and scarcity of labeled data, thereby reducing the accuracy and robustness of the models. In this paper, we proposed an uncertainty- guided adversarial semantic transformation learning algorithm (USTAL) that use semantic transformation to augments the source domain data and we add the MC dropout layer to explicitly construct the model and predict uncertainty. The method dynamically generates samples with significant domain shifts and high uncertainty during the adversarial training process, and enhances the model's adaptability to the unseen target domain through the combination of entropy regularization and contrastive learning. Experiments show that USTAL significantly improves the robustness and accuracy of the classification model in multiple target domains.