Target tracking method for unsupervised similarity discriminant learning

The invention discloses a target tracking method for unsupervised similarity discriminant learning, and relates to the technical field of computer vision target tracking. The method comprises the following steps of performing dimensionality reduction and clustering on features by adopting t-SNE to obtain image category pseudo tags and a category total number, then taking the pseudo tags as image real tags, inputting the image features into a full connection layer and performing classification, and performing back propagation training by a network; after training is completed, the network has an image similarity discrimination capability; in a tracking process, firstly obtaining a first frame image, obtaining a to-be-searched area around a target corresponding to the current input frame according to an intersection ratio IOU greater than 0.8; acquiring target candidate blocks by using particle filtering, acquiring features of the target candidate blocks by using an unsupervised similarity feature extraction la