Night vision image de-noising method by wavelet fuzzy threshold application in apple harvesting robot

• 1、School of Electrical and Information Engineering, Jiangsu University
• 2、Jiangsu Unversity

Abstract：In order to achieve the round-the-clock operation of apple harvesting robot, the relevant studies on the night vision image have been done. Due to the real-time night vision image has very much noise, lead to the quality of the image is decreased, and the operating efficiency and recognition precision of the image processing system of apple harvesting robot are influenced, thus affects the overall harvesting efficiency. On the basis of the analysis of noise, the wavelet theory is introduced into image processing system, using the wavelet threshold method for image noise reduction processing. However, the selection of wavelet threshold function restricts the effect of de-noising. Aiming at the potential disadvantages of threshold algorithm, the fuzzy theory is introduced into wavelet algorithm, through constructing fuzzy threshold functions, and the wavelet fuzzy threshold de-noising algorithm is proposed. The purpose of new method is that, to reduce the noise of the image as much as possible, to achieve performance of enhanced and de-noising of night vision image, and be beneficial to further analysis and recognize. In order to from the objective numerical measure denoising effect of each denoising method, the concept of relative peak signal-to-noise ratio (RPSNR) is proposed. RPSNR consistent with natural light image as standard (as no noise image), it regard as the reference signal of other pending images, and then to calculate the peak signal-to-noise ratio. Several night vision images as unified standard reference to natural light image, the calculation standard is the same, so RPSNR has comparable. In the experiment, night vision apple images are collected under three kinds of artificial light, and use median filter, wavelet soft threshold and wavelet fuzzy threshold method denoise processing. The experimental results show that, from the vision, the low noise images are got by WT-ICA denoising method, is noise reduced sharply; from the RPSNR, relative to original night vision image and wavelet soft threshold denoising method, the new denoising method average increased 19.69%, 9.19%, the advantage of new denoising method is obvious; Comparing several kinds of artificial light, incandescent lamp is suitable for artificial light. They show the unique advantages for the night vision image noise reduction, to lay a solid foundation for the round-the-clock operation of apple harvesting robot.

Keywords： Apple Harvesting Robot Fuzzy Wavelet De-noising Night Vision Image Relative Signal-to-noise Ratio