A soil image represents soil characteristics under the same conditions, which will undoubtedly improve the accuracy of soil species identification. The enhanced soil images may approach the real soil images with certain brightness if brightness of the soil images collected by machine vision in the natural environment can be enhanced controllably. This will eliminate or weaken the effect of the different natural illumination in the future soil species identification. The Sigmoid curve is introduced to fit the cumulative probability density (cdf) curve of the Y component of the soil image. Then, an optimization model of approaching target luminance is established to transfer the fit sigmoid curve and realize the migration of soil image luminance. Next, according to the neighborhood information of the pixels, the pixel with same brightness are sorted and migrated to finish the controllable luminance enhancement of the soil image. Finally, the low-frequency of U and V components are extracted by Gaussian convolution kernel. The low-frequency and high-frequency of U and V components are transformed according to the color ratio invariance and the neighborhood information of the original soil image. Finally, the enhanced Y, U and V components are fused to obtain the enhanced color soil image. The experiments are done with the proposed algorithm based on the image pairs of real soil images under different brightness. Their results show average standard deviation of Y, U , and V component differences of the corresponding pixel between the enhanced soil image and the real target soil image are 14.313 7, 1.323 2, 2.110 5 respectively, And the average peak signal-to-noise ratio is 29.820 9. The average standard deviation obtained by the proposed algorithm is 0.767 7~4.762 9, 0.052 4~1.110 4, 0.071 4~1.272 0 lower than the comparison algorithm 2-D HS, WGSF respectively. It has high precision and low distortion, and its effective brightness enhancement range is ［-35, 35］. These prove that the algorithm is effective.