SE-ResNet-based noise reduction for steady-state micro-thrust measurement

Micro-Newton thrusters are widely utilized in the field of astronautics. Typically, the precision of micro-newton thrust measurement is fundamentally hinged upon the level of background noise. In this research, we introduce Residual Neural Network (ResNet) to identify the effective signals merged in the background noise. Experimental studies are carried out to investigate the effect of noise reduction of ResNet. Squeeze-and-Excitation (SE) block and the SE-ResNet are then adopted to optimize the net. It is shown that steady-state signal with 0.1μN as the minimum change unit can be recovered from the noises with amplitude of 0.8μN, and the accuracy reaches 70.70% with ResNet. Besides, SE-ResNet shows better performance with accuracy of 73.41% than the conventional ResNet. The proposed method has great potential for noise reduction of steady-state sensor signals.