CBGTE: Neural Network Aided Extended Kalman Filter for Dual-Band Infrared Attitude Estimation

Due to the numerous advantages of dual-band infrared radiation (DBIR) attitude measurement (AM) technology, it has garnered significant attention from industry and academia. However, geometric errors caused by sensor measurement noise, assembly positions, motor interference and other sensor-related commonalities, along with random errors induced by the sensitivity of DBIR characteristics to infrared radiation interference, collectively undermine the reliability of the estimated attitude information. To address this issue, the bidirectional gated recurrent unit (Bi-GRU) and Transformer were combined to assist extended kalman filter (EKF) for DBIR attitude estimation (CBGTE). The core concept involves two aspects: 1) utilizing EKF to mitigate geometric errors during the process of DBIR AM, and 2) combining Bi-GRU and Transformer to aid EKF in compensating for random errors in the measurement process. A semi-physical experimental platform is established to validate the performance of CBGTE. Through experimental validation with real-world data, the proposed CBGTE algorithm has demonstrated significant improvements in accuracy when compared with several state-of-the-art algorithms, achieving roll angle error of ±0.4° and pitch angle error of ±0.2°.