MOEA-LoANets: An Online Trajectory Generation Framework for Synthetic Aperture Radar Based on Block Propagation Strategy

Synthetic aperture radar (SAR) is a kind of microwave remote sensing technology, which can work in all-day and all-weather conditions. With an external illuminator, the receiving platform, e.g., airplane and missile, can passively receive the backscattered signal and achieve continuous bistatic SAR imaging of the target scene by adopting a proper flight trajectory. The target information can then be generated from the updating images for advanced applications, such as aircraft self-landing and missile guidance. However, atmospheric disturbance and positioning error will cause the actual trajectory to significantly deviate from the planned trajectory, which deteriorates the imaging and guidance performance. Therefore, in this article, the real-time SAR trajectory planning under external perturbations is investigated. A two-stage online trajectory generation framework considering trajectory control and SAR imaging performance is put forward, namely, multiobjective evolutionary algorithm based on large-scale optimization-based adaptive networks, to generate a flight path according to the real-time state information of the receiving platform. In the first stage, the framework utilizes multiobjective evolutionary algorithms to solve the perturbed trajectory planning problem (TPP) set and generate the datasets for network training. A block propagation strategy is devised to efficiently solve the massive number of TPPs with different initial states under perturbations. Then, multiple datasets are integrated for the training of large-scale optimization-based adaptive networks (LoANets) to obtain a real-time trajectory generator. In the LoANets, a cooperative differential evolution method is proposed and integrated with the regularized extreme learning machine to enhance the stability and generalization performance of the network. The proposed framework can generate a path with optimized bistatic SAR imaging and trajectory control performance under external disturbances. Experimental results demonstrate the effectiveness of the proposed framework.