Graph-Based Spatio-Temporal Semantic Reasoning Model for Anti-Occlusion Infrared Aerial Target Recognition

Infrared target recognition and anti-interference incomplex battlefields is one of the key technologies enablingthe precise strike capability of aircraft. Currently, infrared-guided aircraft face complex interference such as naturalbackgrounds and artificial decoys, leading to a decrease in theperformance of infrared target recognition. A particular challengeto infrared target recognition and anti-interference capabilitiesis the strong interference situation caused by the combination oftarget maneuvering and the dense, continuous, and coordinateddeployment of infrared decoys. To address extreme issues such ascomplete loss of target feature information and inability to identifydue to target occlusion, we develop an anti-interference recognitionmethod based on a visually inspired Spatio-Temporal SemanticReasoning Model (STSRM). Firstly, inspired by the functionalcharacteristics of visual semantic reasoning, the STSRM isproposed to simplify the reasoning of relationships among multipleregions into modeling relationships between corresponding regionnode features in a graph-based module. Secondly, an anti-occlusiontarget recognition model based on STSRM is constructed, whichintroduces a reasoning graph module connecting node regions toinfer semantic information and predict targets between regions.The test results on the infrared dataset established in this paperindicate that the proposed anti-interference recognition model canmake accurate target predictions in large-scale or full-occlusionconditions, and we achieve 13.9% and 3.1% improvement on mAPscores and mIoU scores, compared to current advanced method onour simulated infrared dataset.