Few-shot incremental learning refers to continuous learning from a limited number of samples to classify new classes without forgetting the old ones. In this paper, we propose a few-shot incremental learning method based on style perturbation from adjusting the position of sample features in the feature space. First, the mean and variance of the samples are computed, and subsequently after adversarial attack a perturbation is applied to them to get the mean and variance after the perturbation. The joint original samples are fed into the SP module proposed in this paper to get the post-migration features, which make the model classifier more robust under the effect of the migrated features, modifying the distribution of the sample features in the embedding space so that the intra-class distance is reduced and the inter-class distance is expanded. At the same time, the calibration module is used to adjust and reduce the semantic difference between the new class classifiers and the old classifiers to improve the generalization ability of the model. This paper presents extensive experimental results and analysis on three datasets, all of which achieve excellent performance.