Lightweight Tracking of Satellite Video Object Based on Saliency Enhancement Mechanism

Target tracking based on satellite platforms in remote sensing images plays a critical role in military and civilian fields. However, most of the traditional algorithms are still aimed at natural scenes and are difficult to be directly applied to complex satellite images with large fields of view and weak contrast. Therefore, the method of tracking satellite videos based on a multidimensional enhancement mechanism is proposed. For the problem that the complex background in satellite images, which makes the target difficult to be correctly captured and identified, the triplet attention module is introduced to enhance the significance of the target in an efficient way, thereby improving the performance of the tracking network; because of the large computational complexity of a deep convolution network, the network structure with the ghost feature is adopted, and some traditional convolution operations are replaced by simple linear operations, which improves the speed of the network. Finally, with the support of satellite remote sensing datasets, the effectiveness of this method is verified through qualitative and quantitative experiments.