Research On Volume Measurement And Grading Method Of Red Jujube Based On Machine Vision

Xinjiang is the main producing area of jujube in China,whose planting area and yield account for more than 2/3 of the whole country.Thanks to the special geographical environment and climate conditions,the jujubes produced in Xinjiang are of high quality.However due to the underdeveloped processing and grading technology,the added value of Xinjiang jujube is not enough high and seriously inconsistent with its quality,which is an important factor restricting the development of Xinjiang jujube industry.Grading is an important means to improve the value and quality of jujube.The current grading methods of jujube mainly include manual grading and mechanical grading,of which manual grading is not only inefficient,but also easy to be affected by subjective factors,so that its grading accuracy is not enough high.Mechanical grading is mainly based on the one-dimensional characteristics of jujube such as long diameter and length,and the grading index is relatively single,with large errors.Using machine vision technology to grade jujube can not only improve the grading efficiency and the detection accuracy,but also reduce the detection cost,which is the development trend of jujube grading technology.For this reason,this paper carries out the research on two-dimensional and three-dimensional information measurement methods and grading technology of jujube based on machine vision technology,which provides scientific basis for the selection of grading feature parameters,the formulation of grading standards and the research and development of grading equipment.The main research contents and results are as follows:1.Study on the physical characteristics of different grades of Xinjiang jujube and its influence on the result of volume gradingFirstly,we measured eight characteristic parameters of jujube,including length,long diameter,short diameter,perimeter,contour perimeter,projection area,weight and volume by different methods.Then we conducted statistical analysis of the relationship between characteristic parameters and grade,and respectively drew the box plot between different characteristic parameters and grade,the linear fitting figure between characteristic parameter mean value and grade.The results showed that there were significant differences between the characteristic parameters and grades of different grades of jujube,and there was a significant linear correlation between the characteristic parameters and grades of different grades of jujube.Secondly,we conducted the correlation analysis and cluster analysis of eight characteristic parameters of jujube and found that the correlation between projection area and contour perimeter was the highest,r=0.96;the next was the correlation between length and contour perimeter,r=0.79.In addition,the correlation coefficients between short diameter and perimeter,length and projected area,volume and projected area,volume and contour perimeter were all between 0.74 and 0.75,while the correlation between other characteristic parameters was relatively small.The cluster analysis of eight characteristic parameters of jujube showed that the characteristic parameters could be divided into five categories.The 1st category was perimeter and short diameter;the 2nd category was length diameter;the 3rd category was volume;the 4th category was length;and the 5th class was weight,projected area and contour perimeter.Thirdly,we analyzed the errors between the long diameter and the short diameter,the long diameter and the length of different grades of jujube,and the correlation between the characteristic parameters and the volume of jujube.And we sorted the other characteristic parameters based on the keywords of the volume characteristic parameters to calculate the error ratio of other characteristic parameters.The results showed that the correlation between the long diameter and the volume was the best,whose correlation coefficient r was 0.9550.The correlation between short diameter and volume was the lowest,whose correlation coefficient r was 0.8613.The error rate of length was the highest,which was 42.2%,and the average error rate of long diameter was the lowest,which was 25.6%.2.This paper researched and put forward the volume measurement method and grade determination technology of jujube based on 2D imageWe collected the two-dimensional image of jujube,and extracted five characteristic parameters of jujube,including its length,long diameter,short diameter,contour perimeter and projection area by image processing method.Then based on the jujube volume we measured,we conducted linear fitting between the characteristic parameters of each jujube image and the volume.The results showed that there were significant linear correlations between the five characteristic parameters and the jujube volume,of which the correlation between projection area and volume was the highest,with the determination coefficient R2=0.9412.Then we used the univariate linear regression model of projection area and volume grade to conduct volume grading.The highest accuracy of jujube volume grade prediction was 95%.We used the principal component analysis method to screen these five characteristics again and selected the first two principal components to build the multivariate linear regression model of jujube volume grade.The highest accuracy of the model on judging the jujube volume grade was 100%,and the lowest accurary was 80%.The experimental results showed that the volume of jujube can be predicted by a single feature parameter or a combination of multiple feature parameters.3.This paper studied the 3D modeling technology of jujube based on 2D image features and put forward a fast volume measurement method of jujube based on 3D model to determine the jujube gradeAfter collecting image of jujube samples of different grade,we extracted the characteristics of jujube image such as the length,long diameter and short diameter,and established three kinds of 3D models of jujube volume,which were ellipsoid,cylindroid and elliptical cone.We used this model to predict the volume of jujube and determine the volume grade of jujube.It was found that among the three volume models,the average relative error and root mean square error of different grades of jujube calculated by ellipsoid model were the smallest;and the accuracy of the jujube volume grade determined by ellipsoid model was the highest,which reached 100%.Experiments showed that 3D model were established by using 2D image features,and 3D model were used to measure the volume of jujube.4.This paper proposed a fast measurement method of jujube volume and surface area based on multi contour 3D reconstructionWe set up the 3600 image collection device of jujube by using industrial camera and rotating disc,and collected jujube image with an interval of 15°,12°,9°,6° and 4°respectively.Then based on the conversion relationship between two-dimensional contour and three-dimensional space we derived,we transformed the two-dimensional contour of jujube into three-dimensional multi contour point cloud model.And we studied the influence of different projection height,included angle and diameter of contour on the calculation accuracy of model volume and surface area.The results showed that when the diameter was fixed,the relative error of jujube volume calculated by the multi contour model would increase with the increase of contour included angle and projection height,and the relative error of surface area would decrease with the increase of contour included angle and projection height.When the contour included angle and projection height of the multi contour model were fixed,the relative error of volume and surface area would change significantly with the diameter.If the multi contour volume model of jujube was used to determine the volume grade of jujube,the highest accuracy was 100%and the lowest accuracy was 70%.The results showed that the method of volume and surface area of jujube based on multi contour 3D point cloud model was feasible,which has a certain reference value for jujube grade.5.This paper carried out research and grading test on the method of measuring the volume and surface area of jujube based on 3D point cloud imageFirstly,we fixed the jujube on the special fixture,collected the point cloud images of four sides of the jujube with line laser scanner,and obtained the three-dimensional point cloud model of jujube through registration.After processing the point cloud model with filtering,smoothing,bug repairing,down sampling,coordinate transformation and translation transformation,we measured the volume and surface area of jujube point cloud model by convex hull method,projection slice method and triangular mesh method respectively,and carried out the grading test.The results showed that when using the point cloud model to measure the volume and surface area of the standard sphere with a diameter of 19.85mm,35mm and 40.4mm,the measurement error of triangle mesh method was the largest,then the projection slice method,and the convex hull method at last.The surface area error of projection slice method was the largest,while the errors of the triangle mesh method and the convex hull method were smaller.At the same time,when using these three methods to measure the volume and surface area of the sphere,the measurement error would increase with the increase of the sampling interval,and the sampling time would decrease with the increase of the sampling interval.When using the convex hull method to measure the volume and surface area of the jujube point cloud model and determine the jujube volume grade,the highest accuracy was 90%and the lowest accuracy was 60.0%.The above three methods can provide technical reference for automatic measurement of three-dimensional characteristic parameters of jujube and development of grading equipment.6.Design and development of online grading system for red jujube volumeThis paper studied the instantaneity and feasibility of the online jujube volume grading system based on machine vision,and developed a set of jujube volume online grading system composed of industrial camera,PCL,belt conveyor and pneumatic device based on the jujube volume prediction model we developed.We tested and evaluated the jujube volume grading method of this platform,and tested the grading accuracy of this jujube volume grading system at different speeds.The test results showed that the grading accuracy of jujube volume grading was 80%,88%and 97%respectively at V1=0.2m/s;V2=0.18m/s and V3=0.13m/s,indicating that the higher the grading speed,the lower the grading accuracy.The combination of machine vision technology could achieve the grading and prediction of jujube volume.