Research On Defect Detection And Classification Of Jujube Based On Deep Learning

Jujube has high nutritional value and planting jujube trees has brought good economic effects to remote areas.In recent years,jujube industry has developed rapidly and has become a pillar industry in many remote areas.Red jujube,before entering the market of defect on the separation of the out of date,quality and the size of the red jujube grading still is given priority to with artificial way,the intensity of labor is larger,there are also some areas have begun to adopt mechanical automation red dates in the form of a preliminary classification,because the artificial classification and the classification accuracy and efficiency of mechanical classification,the commercialization of jujube is uneven in quality,directly affects the red jujube after processing.At present,the varieties and quantity of deeply processed jujube products are less,and the price and added value of jujube are generally lower.Aiming at the problem of grading accuracy and efficiency of jujube,this subject adopts machine vision fusion deep learning technology to detect and grade the surface quality of jujube,improve the grading efficiency and commercial added value of jujube,and promote the rapid development of jujube industry.In this study,the phenotypic characteristics and surface defects of dried ruoqiang jujube were detected by machine vision.The main research contents of this paper are as follows:(1)collect jujube samples and measure the phenotype and geometric parameters of jujube.On the basis of LY/T 1780-2018 "quality grade of dried red jujube" industry standard,the quality standard of ruoqiang jujube was determined,and the size grading standard of ruoqiang jujube was established according to the shape characteristics of ruoqiang jujube.(2)The overall design of jujube testing and grading machine was put forward,and the feeding and conveying system,transmission system,visual inspection system and electromagnet grading system of the equipment were studied and analyzed.Through the design and calculation of each motion mechanism of grader,the parameters of each motion mechanism are determined,and the 3d model of jujube detection grader is established.(3)The method of deep learning based on machine vision was proposed to detect the surface defects of jujube.Inthis paper,the surface defects of jujube were detected by improving the traditional convolutional neural network.Then,the structure and parameters of the traditional deep residual network are improved,the defect detection model is compared with the convolutional neural network model,and the surface defect detection model of jujube based on the improved residual network is established.Experimental results show that the accuracy of defect recognition in test samples can reach 96.11%when the depth residual network model is adopted.(4)Proposed to ovalization processing get red jujube appearance to elliptic long axis and short axis,if qiang grey jujube grading standards promulgated by the combination,the long axis length instead of red jujube jujube size grading,build the regression relationship between length of red jujube and jujube weight,and calculated the area of the normal red jujube to ovalization and the difference between the red jujube outline area,determine the scope of the error,get deformity fruit judgement and grading standard.The experimental results show that the correlation coefficient between the length and weight of jujube is 0.7468,and the error range between the normal jujube contour area and the quasi-elliptic area is no more than 27.3mm2.(5)According to the actual production needs,the function modules of the hierarchical detection system are divided by means of modular design.Qt designer software for the red jujube classifier software interface design,and then Qt and Spyder software mixed programming to achieve the control of the software interface function.