Single target tracking method based on example attention mechanism

The invention discloses a single target tracking method based on an example attention mechanism. The method comprises the following steps: S1, obtaining a deep fusion feature map of a template image and a search image; s2, calculating instance-level self-attention of the deep fusion feature map, and obtaining a response map; and S3, performing target positioning and bounding box regression according to the obtained response diagram. According to the method, a popular attention mechanism and a twin network structure in the current visual target tracking field are combined, meanwhile, a lightweight backbone network is adopted, template features and search features are fully fused through a pixel-level feature fusion module, a multi-channel feature map is abstracted into feature vectors by using adaptive maximum pooling and adaptive average pooling, and the feature vectors are extracted into the target tracking algorithm. And data set information is packaged into example representation, and information loss caus