In the encrypted network traffic intrusion detection, deep learning based schemes have attracted lots of attention. However, in real-world scenarios, data is often insufficient (few-shot), which leads to various deviations between the models prediction and the ground truth. Consequently, downstream tasks such as unknown attack detection based on few-shot will be limited by insufficient data. In this paper, we propose a novel unknown attack detection method based on Intra Categories Generation in Embedding Space, namely SFE-GACN, which might be the solution of few-shot problem. Concretely, we first proposed Session Feature Embedding (SFE) to summarize the context of sessions (session is the basic granularity of network traffic), bring the insufficient data to the pre-trained embedding space. In this way, we achieve the goal of preliminary information extension in the few-shot case. Second, we further propose the Generative Adversarial Cooperative Network (GACN), which improves the conventional Generative Adversarial Network by supervising the generated sample to avoid falling into similar categories, and thus enables samples to generate intra categories. Our proposed SFE-GACN can accurately generate session samples in the case of few-shot, and ensure the difference between categories during data augmentation. The detection results show that, compared to the state-of-the-art method, the average TPR is 8.38% higher, and the average FPR is 12.77% lower. In addition, we evaluated the graphics generation capabilities of GACN on the graphics dataset, the result shows our proposed GACN can be popularized for generating easy-confused multi-categories graphics.