Unsupervised vehicle re-identification method based on multi-semantic knowledge

The invention provides a method for carrying out cross-domain unsupervised vehicle re-identification by utilizing multi-semantic knowledge, which specifically comprises the following steps of: proposing to utilize a multi-semantic knowledge learning method to analyze the similarity between vehicle pictures, and obtaining multi-semantic information by Focal Drop Network; and clustering analysis is carried out on the semantic information to determine various labels and the labels are used for training a vehicle re-identification model. In order to improve the accuracy of the label, a triple center loss function is provided for the difference between the intra-class distance and the inter-class distance, and the generalization ability of the model is improved to adapt to an unknown domain. A comprehensive experiment result shows that the method disclosed by the invention has good performance on both a VehileID data set and a VeRi-776 data set. 本发明提供一种本发明公开了一种利用多语义知识进行跨域无监督车辆再识别的方法，具体地，包括：提出利用多语义知识学习方法来分析车辆图片之间的相