Through the analysis of noise, it’s found that the Gaussian noise is the main noise in the night vision images obtained under different artificial lights, which also mixed with some salt and pepper noises. With regard of the elimination of Gaussian noise, the wavelet transform (WT) and independent component analysis (ICA) were introduced into the process of night vision images. In order to minimize the noise, a combined method of WT and ICA (WT-ICA) was proposed. The simulation results verified the effect of this combined method. For the purpose of better evaluation of denosing effect in these night vision images, taking the image under the natural light as a reference, an index named relative peak signaltonoise ratio (RPSNR) was proposed. The repeated tests were carried out in different night vision images. The results showed that there was an obviously visual decrease of noise with WT-ICA method. The RPSNRs of WT-ICA images were improved by 29.94%, 8.09% and 7.54% than those of original images, wavelet soft threshold denoising images and ICA denoising images. Especially under the incandescent lamp, the RPSNR reached the highest value, so this kind of lamp was suitable for being artificial light. By means of continued processing with WT-ICA method, these low noise images were easy to be identified further, which laid a solid foundation for the roundtheclock operation of the apple harvesting robot.