Small sample target detection method based on sample semantic correction in knowledge graph

The invention discloses a small sample target detection method based on sample semantic correction in a knowledge graph, and belongs to the field of computer vision, target detection and small sample learning. In the prior art, in a fine tuning process of a small sample target detection method, semantic information of a new class needs to be established on the basis of base class knowledge. In the process, the new class is inevitably close to the similar base class in the decision space, so that misjudgment occurs during final decision making of the network. According to the method, the semantic relationship between the categories is established by trying to establish the semantic relationship between the categories through the knowledge graph, and the new class offset in the decision space caused by the scarcity of the samples is corrected, so that the problem that a small sample detector cannot accurately identify the samples is solved. 该发明公开了一种基于知识图谱中样本语义校正的小样本目标检测方法，属于计算机视觉、目标检测和小样本学习领域。针对现有技术在小样本目标检测方法的微