The images of locust slices with abundant textures are often negatively impacted by external noises during image acquisition, enhancement and so on. These noises then destroyed the textures of the locust slice images and hindered the study of the locust cell structures. In the frequency domain, traditional 2D wavelet transform based on tensor, in which only have two directions, horizontal and vertical, can’t deal with high dimensional data effectively. Besides, the filter that constructed wavelet was isotropic, so the traditional denoising methods using wavelet denoising noise made image edge and texture blur. Shearlet often uses a compactly supported traditional wavelet function as its wavelet basis. Then this wavelet basis though translation, dilation and shear transform, makes shearlet express multiple directions. The shearlet filter is anisotropic. Shearlet uses the special frame structure, in which can preserve texture and edge so as it will not be noised. Shearlet transform was based on multiscale geometric analysis. Shearlet can represent image sparsely. Thus this paper proposed a Shearlet algorithm based on Meyer window function. The algorithm used Meyer wavelet as the wavelet basis function, since Meyer wavelet was symmetric and infinitely differentiate. First, Meyer wavelet function and scale function were used to construct a Meyer window function. 〖HJ〗Meyer window function was used to decompose the noisy locust slice images in frequency domain, and compute Shearlet norms in each scale and each direction. Second, the traditional hard threshold method was used to process the Shearlet coefficient. Finally, through Shearlet inverse transform, the locust slices images were restructured. In the experimental section, two groups of experiments were set up. In the first group experiment, the proposed algorithm was compared to other denoising algorithms, such as Meyer wavelet threshold denoising algorithm, partial differential equation denoising algorithm and so on. The classical image quality evaluation index was adopted to evaluate the performances of these algorithms. The adopted evaluation index included mean squared error (MSE), peak signal to noise ratio (PSNR), and structure similarity (SSIM). When the noise standard deviation was 30, the PSNR got by this paper was 2.5dB higher than that got by Meyer wavelet threshold denoising algorithm, and 2dB higher than that got by partial differential equation denoising algorithm. The second group experiment, in order to verify the retain texture with different directions, a locust cell image with much more texture was selected. The final experiment result showed that this algorithm can retain effectively. Others traditional denoising algorithm made the edge and textures blur.