In order to better obtain the edge details and texture information of the image in the process of denoising, the balance was obtained in the removal and excessive smoothing of the image detail noise. A parameterize threshold function for denoising algorithm was proposed based on Bandelet transform, which took full advantage of the multiscale characteristics of Bandelet transform and the geometric characteristics of image.The image was decomposed by using stationary wavelet with translation invariance to overcome the oscillation of the image and the threshold was estimated by Birge-Massart strategy. Then the optimal geometric flow direction was obtained by minimizing Lagrange function. The quadtree of Bandelet transform was optimized according to minimum mean square error (MSE) principle. Finally, the adaptive Bayesshrink parameterized threshold function was used to image denoising. The results showed that the proposed method performed more effectively to preserve the edge features and the fine structure of the denoising image. Compared with other methods, the peak signaltonoise ratio (PSNR) and structural similarity (SSIM) obtained by the proposed algorithm showed that the performance of noise reduction was improved significantly. Therefore, the parameterized threshold function denoising algorithm based on Bandelet transform was feasible and effective in image denoising of locust slices, which provided technical support for its subsequent processing.