Improving the accuracy of the kalman filter algorithm in AHRS sensor using LSTM deep learning module

Accurate Attitude and Heading Reference System (AHRS) play an essential role in navigation and guidance Unmanned vehicles. Today, the use of various algorithms and methods, including adaptive filter, neural network and predictor filters, to increase the accuracy of these systems and reduce the noise of its sensors, has received much attention from researchers. In this paper, the combination of LSTM deep neural network and kalman filter is used to improve the accuracy of AHRS system. In this method, first the deep network used is trained and then as a corrector, it corrects the effective coefficients of Filter Kalman. This method removes all the limitations of Kalman filter, including its linearity and non-memory, and improves the accuracy of the output attitudewithout the use of GPS. The experiments in this study, are based on real low-cost MEMS-based IMU sensors, which is less accurate than sensors used in similar article, installed on high-maneuverability UAV, present about 35% accuracy improvement in attitude estimation and 40% reduction of the output noise.