Rotary attention single target tracking method and device based on deep neural network

The invention discloses a rotation attention single target tracking method and device based on a deep neural network. The method comprises the following steps: extracting features of a current frame image and features of a template image comprising a tracking target; fusing the extracted features based on a rotation attention module, and extracting the features of the current frame image again through a preset decoder; based on a sparse attention module, the features fused based on the rotation attention module and the features, extracted again, of the current frame image are fused again, and target features are obtained; performing classification regression on the target features to obtain a classification result and a regression result; and determining the position of the tracking target in the current frame image according to the classification result and the regression result, thereby realizing high-precision tracking of a single target in extreme environments of partial shielding, complete shielding, ill