The Research On Shape And Quality Test Grading Of Batch Potatoes Based On Machine Vision

Potato is an important food crop.Before deep processing,potatoes of different shapes and qualities will be graded.At present,roller machinery is widely used for rough sorting,which is easy to cause mechanical damage.Based on the research status at home and abroad,a rapid nondestructive testing classification method is proposed,which uses machine vision technology to classify the shape and quality of batch potato targets.The images of batch potatoes are processed,the characteristic parameters are extracted,and the detection and grading of shape and quality are completed.The main research contents are as follows:(1)Compare different models of RGBD cameras and select the camera that meets the acquisition conditions as the acquisition equipment.The imaging principle and distortion principle of Kinect v2 camera are studied,the mathematical model is established.According to the marking principle of color lens and depth lens,Zhang Zhengyou calibration method is used to calibrate the color lens and depth lens respectively,and finally the camera parameters are solved.(2)For the two-dimensional potato image,the median filter method is used for denoising,and then the iterative method is used for threshold segmentation and morphological closed operation processing.Then the Laplace operator and Gaussian operator are combined to detect the potato edge.In view of the phenomenon of image occlusion and missing in shooting batch potato images,the watershed algorithm is applied to segment independent potato individuals,then the missing edge texture is synthesized according to the known area,and the missing image is predicted and repaired based on curvature characteristics.Finally,the rectangularity,the length width ratio of the minimum circumscribed rectangle and the roundness parameters of the batch potato are displayed.(3)Firstly,the point cloud images from different perspectives are segmented effectively by using the segmentation algorithm based on improved curvature constraint,and then the outlier is removed by using the outlier elimination method based on the point cloud radius.Then the forward normal vector is solved,and the improved iterative nearest point algorithm is used to register the point cloud images from different perspectives,and optimize the registration.Repair the holes with point clouds,and then use Poisson reconstruction to complete the establishment of three-dimensional model.Based on the cluster triangulation technology of another stronghold,the Delaunay triangular mesh model of potato surface is established by using Lawson algorithm based on optimization criteria,and then the tetrahedral cone model is obtained.The mass is obtained by multiplying the cone volume in the calculation space by the potato density,and the triangular mesh model of batch potato targets is established.(4)The shape and quality of batch potatoes were tested.The broken line diagrams of parameters of round potato,oval potato and deformed potato were obtained,the change law was analyzed,and the shape was judged.The total accuracy was 95.7%;The detection data of multi view point cloud reconstruction technology are obtained by experiment,and the actual quality is analyzed by linear regression;The results show that this method has less error and faster detection speed,and can grade the quality of batch potatoes.