This paper presents a comprehensive survey of recent advances and challenges in few-shot object detection (FSOD). FSOD aims to detect novel objects with limited annotated samples by combining few-shot learning and object detection techniques. The paper introduces a novel FSOD taxonomy based on transfer learning principles, categorizing methods into episode-task-based and single-task-based approaches. Episode-task-based methods use meta-learning to adapt to novel tasks through multiple episodes, while single-task-based methods transfer pre-trained base model parameters to the novel stage for fine-tuning. The paper discusses the advantages and limitations of these methods, highlighting challenges such as class imbalance, catastrophic forgetting, and interpretability issues. It also reviews key FSOD algorithms, benchmark datasets, and evaluation protocols, including Pascal VOC, MS COCO, LVIS, and FSOD. The paper concludes that FSOD methods based on transfer learning can effectively adapt to data-scarcity scenarios, but further research is needed to address challenges like feature reweighting, interclass relationships, and metric learning. The survey provides a comprehensive overview of FSOD research, offering insights into current trends and future directions in object detection.This paper presents a comprehensive survey of recent advances and challenges in few-shot object detection (FSOD). FSOD aims to detect novel objects with limited annotated samples by combining few-shot learning and object detection techniques. The paper introduces a novel FSOD taxonomy based on transfer learning principles, categorizing methods into episode-task-based and single-task-based approaches. Episode-task-based methods use meta-learning to adapt to novel tasks through multiple episodes, while single-task-based methods transfer pre-trained base model parameters to the novel stage for fine-tuning. The paper discusses the advantages and limitations of these methods, highlighting challenges such as class imbalance, catastrophic forgetting, and interpretability issues. It also reviews key FSOD algorithms, benchmark datasets, and evaluation protocols, including Pascal VOC, MS COCO, LVIS, and FSOD. The paper concludes that FSOD methods based on transfer learning can effectively adapt to data-scarcity scenarios, but further research is needed to address challenges like feature reweighting, interclass relationships, and metric learning. The survey provides a comprehensive overview of FSOD research, offering insights into current trends and future directions in object detection.