Deep learning technology has achieved remarkable success in object detection, finding extensive application in areas such as traffic management and obstacle avoidance. However, practical applications often face the challenge of acquiring a large volume of sample data. Consequently, the academic community has recently shifted its focus towards algorithms capable of learning from a small number of samples. This article aims to review the main few-shot learning methods and their applications in object detection tasks. It summarizes the strengths and limitations of metric models, memory models, parameter update models, and sample augmentation models, discussing their real-world applications in object detection. Additionally, the article analyzes the limitations of few-shot learning methods and looks forward to future research directions in object detection based on few-shot learning, focusing on reducing data dependence, improving algorithm efficiency, and enhancing model adaptability.