A pruned wavelet neural network noise optimization method for fiber optic gyroscope

Abstract The fiber optic gyroscope is a precision instrument for inertial guidance. In order to enhance the noise-reduction capabilities of compact fiber optic gyroscopes and address the challenge of limited logic resources in migrating and transplanting wavelet neural networks to FPGA, this paper proposes a wavelet neural network approach based on the average percentage of activation neuron pruning to achieve noise optimization of fiber optic gyroscopes within the constraints of limited FPGA logic resources. The wavelet neural network method is employed to optimize the noise-filtering effect. The network complexity is reduced by monitoring the neuron output as a criterion for neuron pruning. Following the implementation of the optimization method proposed in this paper, an average reduction of 20% in algorithmic logical resources was found. Compared with the third-order wavelet denoising, the output variance of this method is 0.503 times that of wavelet denoising. This method has engineering significance for the miniaturization and integration of fiber optic gyro systems.