Research of night vision image denoising method based on the improved FastICA

When apple harvesting robot operates at night, there are lots of noise in the apple's night vision image captured by image processing system. And the noise is mainly Gaussian noise, and mixed with some Salt and Pepper noise. In order to improve the harvesting efficiency and precision, the Independent Component Analysis (ICA) is introduced into the denoising method for night vision image. Mixture ICA model with noise based on the Maximum Likelihood Estimation regards likelihood of the observed signals as the objective function, which leads to the estimation to the mixed matrix poor. Aiming at the problem, this paper improves the objective function by bias removal technique, which can reduce the bias caused by the noise. And ICA is optimized by Fixed-point algorithm, so that the denoising method can operate efficiently. Through ICA transform, the robot image processing system gets the image's independent components, and then these components with noise-free are estimated by using certain denoising method. Finally, this algorithm realizes the night vision image denoising. In order to verify the effectiveness of the improved FastICA, both Median-Average Filtering denoising and FastICA algorithm based on Maximum Likelihood estimation are simulated by MATLAB to compare the denoising effect. From the Relative Peak Signal-to-Noise Ratio (RPSNR), the improved FastICA denoising method in this paper relative to the other two denoising methods, respectively increases by 19.32%, 4.65%. In conclusion, the improved FastICA algorithm has unique advantage for night vision image denoising, which provides a solid foundation for the night operation of apple harvesting robot.