The volume of aviation gamma spectrum data is immense. If only a central processing unit (CPU) is used for data post-processing, it would be constrained by computational efficiency.

This study aims to propose a CUDA-based graphics processing unit (GPU) parallel solution that optimally accelerates the denoising of airborne gamma-ray spectral data using wavelet transformation.

First, the impact of different block sizes on computational time was tested to determine the optimal block size for processing airborne gamma-ray spectral data. Subsequently, a GPU, instead of a CPU, was used to calculate the acceleration ratio for handling airborne gamma-ray spectral data of different volumes, and wavelet basis functions were used for those with the same data volume. Finally, by introducing white noise to the experimentally measured airborne gamma-ray spectral data, the signal-to-noise ratio of denoised data was calculated to optimize the threshold denoising method suitable for parallel acceleration of the GPU.

The optimal two-dimensional block sizes for denoising airborne gamma-ray spectral data are 64×64 and 128×128. Among the wavelet basis functions, those that achieved a total time acceleration ratio exceeding 100 compared to CPU processing account for 80%, while those that reached an acceleration ratio exceeding 90 constitute 91%. The coif5 function achieves an acceleration ratio of 353 times whilst the acceleration ratio of the threshold denoising function approaches 570.

All wavelet functions exhibit insufficient denoising effects at low signal-to-noise ratios and excessive denoising effects at high signal-to-noise ratios. Significant denoising can be achieved using hard thresholding of coif5, soft thresholding of coif1, and improved thresholding of bior3.7.