Automatic Target Detection with Monostatic-Bistatic MiniSAR and Infrared Sensor Data Fusion

Teamwork between monostatic-bistatic miniature synthetic aperture radar (MiniSAR) systems and infrared (IR) sensors mounted on spatially separated drones would bring enhanced trajectory flexibility and anti-stealth/anti-jamming capability. Leveraging the monostatic-bistatic SAR images and the IR images over the same scene of interest, this paper presents a novel framework for multimodal data fusion in automatic target detection (ATD). For monostatic SAR ATD, the YOLOv5 is used as the base structure for the multiview SAR DNN (MSAR-DNN), and the optimum hyperparameters are obtained via evolution. For the more challenging bistatic SAR ATD, a two-stage procedure is followed. In Stage-1, a YOLO variant is employed for target detection. In stage-2, target classification is carried out with an 8-key-sector-view feature fusion mechanism. The multi-view IR DNN (MIR-DNN) features confidence score ensemble based on the unbalanced information provided in diverse-aspect images, for which greater weights are assigned to the images with finer ground resolution. Simulation results show that the proposed method is effective in combining the key multiview target features reflected by the SAR and the IR imagery.