GADF-FIRViT: A Lightweight Interference Classifier for Infrared Attitude Measurement

Dual-band infrared radiation (DBIR) attitude measurement (AM) technology takes DBIR sensors to extract navigational information from infrared radiation fields in different regions of earth. Different radiation interference have distinct impacts on sensors output. Accurate interference identification is crucial for obtaining high precision attitude angle information. However, there is currently a lack of lightweight and high accuracy interference classifier. To address this issue, we propose a DBIR AM interference classification method, which is based on fusion inverted residuals and vision Transformer (FIRViT) using Gramian angular difference field (GADF). First, through the analysis of the DBIR AM system, the semi-physical experimental platform is established. The DBIR data under various radiation backgrounds are collected on the semi-physical platform and preprocessed. Then the data is encoded into 2D-images using GADF to highlight the feature differences between various interferences, and a detailed analysis of the differential features is performed on the encoded images. Next, a high-precision and lightweight model called FIRViT is established, which includes feature extractor and classifier. In the feature extractor, the design of IRMerB can effectively reduce computational complexity while maintaining accuracy. GADF images are input as training and testing to achieve interference classification. The results show that GADF effectively highlights differences between various interferences, and the FIRViT with GADF is proved with excellent recognition performance and lightweight, reaching an average classification accuracy of 99.8% with 658.53k parameters. This classifier can further strengthen the robustness of DBIR AM technology.