In many application scenarios， the small number of samples imposes large challenges for deep neural networks（DNNs） and few-shot learning（FSL） has received widespread attention in recent years. However， there is a lack of comprehensive and systematic domestic review on this issue. This paper reviews data-augmentation-based and network-model-based approaches of DNNs. They are two key components of the algorithmic framework. Further，representative algorithms of each approach are elaborated. Finally， this paper summarizes current challenges for FSL and future development of FSL. It is expected to provide inspirations for the subsequent research works in this field.