Restoration Of Degraded Images In Potato Grading System

In recent years,machine vision-based application has ushered in unprecedented development.The automatic rapid non-destructive testing system for potato(herein referred to as potato grading system)is an important application of machine vision technology in agricultural automation.During the image collection process of potato grading,the camera and the potato moving on the roller will produce relative motion,which will cause motion blur and introduce noise.Therefore,it is necessary to improve the quality of the degraded image by certain methods before grading,thereby improving the accuracy and efficiency of grading.This paper designs a set of restoration algorithms to achieve degraded images in the potato grading system.The main research contents are:(1)Determination of the blurred image.In order to avoid the problems of the computer to reduce the quality of the image processing and waste the information processing resources,this paper proposes to apply the image before restoring the image.The method of calculating the blurring evaluation parameter(BIM value)is used to determine whether the image is to be deblurred,thereby ensuring the real-time and high efficiency of the potato grading system.(2)Preprocessing of images.Potato images inevitably generate noise during the process of acquisition and transmission,so this paper needs to do some pre-processing before deblurring the blurred image of potato.By comparison,this paper finally chooses the combination of adaptive median filtering and shock filtering to remove the noise and enhance the edges and details of the image that has been destroyed.It lays the foundation for the subsequent more accurate estimation of image blur kernel.(3)Estimation of blurring kernel.Before restoring a blurred image,it needs to estimate the blurring kernel and then apply the corresponding calculation.Firstly,the algorithm of blurring kernel estimation is studied from the airspace and the frequency domain respectively.What's more,the minimum direction differential method and the differential autocorrelation method based on spatial domain are selected to estimate the blur angle and scale of the image respectively.The experimental results show that the estimated error is between 1 and 3 unit values,which is within the allowable range.(4)Restore the image.Based on the research of inverse filtering,Wiener filtering,constrained least squares and RL algorithm,this paper improves the traditional RL algorithm,uses the improved RL algorithm to recover the blurred image,finally,when the roller speed is 70 ft/min,the recovery experiment of the real shot image in the potato grading system is carried out.Experiments on blurring potato images under three different exposure times show that the whole algorithm of this paper has certain applicability and stability.(5)Restoration evaluation.In this paper,the evaluation of the image is interspersed in each part of the above process,which provides an intuitive and objective evaluation criterion for the restoration effect of the algorithm.