To address the issue of significant noise present in the acceleration and angular velocity output signals measured by MEMS sensors, a denoising method based on the pelican optimization algorithm (POA) of variational mode decomposition (VMD) and wavelet thresholding (WT) is proposed. Firstly, POA is used to optimally select the parameter combination of the VMD, then POA-VMD is applied to adaptively and non-recursively decompose the noisy signal into a series of intrinsic modal functions (IMF). Secondly, the IMFs are classified by calculating the cosine similarity of each IMF. Based on the result of the calculation, IMFs are classified into noise-dominant component and signal-dominated component. After classification, the noise-dominated component is subjected to improved wavelet threshold denoising, and finally the processed noise-dominated component is reconstructed with the signal-dominated component to obtain the noise-reduced MEMS sensor signal. The static and dynamic experimental results show that the signal-;to-noise ratio of the denoised signal is improved by 12 and 10 dB respectively, and the mean square error is reduced by 75.5% and 46.6% respectively, which is a significant denoising effect and can improve the accuracy of the MEMS sensor.