Target tracking method, system and equipment based on video SAR

The invention discloses a target tracking method, system and equipment based on a video SAR (Synthetic Aperture Radar), and the method comprises the steps: extracting target semantic features in an image through employing an OfficientNet network, and constructing a multi-scale feature pyramid; performing category prediction and position regression on the anchor point frame on different scale feature maps by using an RPN network; and performing time sequence modeling on the target trajectory by using a Bi-LSTM network, and synchronously prolonging, merging or generating the trajectory of the target on the current frame in combination with the result of the target detection frame. Finally, track information is initialized, and center coordinates of all correct detection frames and misinformation detection frames in each frame are represented; and target center coordinate prediction is carried out based on the Bi-LSTM network to obtain a prediction target. According to the method, semantic information in a high-