Research On Apple Grading Robot System Based On Dual Features

China is a major producer of apples,but the selling price of apples has always been very low.The main reason is that the post-production commercialization of apples is backward,and the appearance quality is poor,resulting in insufficient market competitiveness.Through apple grading,apples with perfect appearance and consistent size can be screened out,to improve the quality of products,reduce the defective rate of products,and improve the market competitiveness of products.At present,the research on apple grading equipment in China mostly focuses on the classification of a single feature,and the grading equipment with multiple features is mostly realized by a combination of different feature grading modules,and the information processing in the grading process is slow.When grading devices grade apples,they mostly use a single queue to detect them one by one,and the grading efficiency is low.In view of the above industry status,this study proposes a grading system that simultaneously grads multiple apples considering the dual characteristics of fruit shape and fruit diameter,to improve the efficiency of apple grading and reduce the labor intensity of workers.The main research contents of this paper are as follows:(1)The Fuji apples produced in North China were selected as the research object,and the apples were classified according to the two characteristics of fruit shape and fruit diameter.Based on national standards and industry standards,the fruit shape and fruit diameter grading indicators were determined.Based on Zhang’s calibration method,the binocular camera required for apple grading is calibrated,and the internal parameter matrix,distortion coefficient matrix and external parameter matrix are constructed to eliminate camera distortion and determine the positional relationship between the left and right cameras.(2)Aiming at the problem of single grading features,based on binocular stereo vision,an improved SSD neural network is applied to build an apple grading algorithm that considers both fruit shape and fruit diameter.Use the color image and depth information of the top of the apple collected by the binocular stereo camera;improve the input layer on the basis of the original SSD convolutional network,integrate the color feature and the depth feature,and realize the simultaneous classification of fruit shape and fruit diameter;for grading In order to solve the problem of slow information processing in the process,the apple classification algorithm was improved by using depth separable convolution,and some ordinary convolutions in the feature extraction network were replaced by depth separable convolution,which reduced the amount of computing data;using The test set evaluates the model and conducts comparative experiments with commonly used convolutional neural networks.The test results show that the improved SSD convolutional neural network algorithm has an accuracy rate of 98.37%on the test set,and the time required for each classification is only 15.96ms under CPU conditions,which improves the information processing speed.(3)Aiming at the problem of low classification efficiency of grading equipment,a robot path planning control algorithm based on greedy rules is designed to improve the sorting efficiency of the sorting system.The algorithm calculates the sorting distance of each apple at the current moment and plans the sorting sequence of apples in real time.The greedy rule algorithm and the sequential sorting algorithm are compared and tested.The test results show that the path planning algorithm based on the greedy rule can effectively shorten the sorting distance of the robot and reduce the sorting time.(4)On the basis of the above work,build a prototype of apple grading.The hardware system of the prototype was built,and the transformation relationship between the camera coordinate system and the robot coordinate system was established through hand-eye calibration;Based on ROS,the software system is built,and each part of the prototype hardware system is controlled to realize Apple’s identification,grading,and sorting.Test verification of prototypes.The test results show that the positioning error of the prototype meets the accuracy requirements,the grading qualification rate reaches 96.9%,which meets the requirements of use,and the sorting time of using the sorting planning algorithm based on greedy rules is reduced by 16.8%,which improves the sorting efficiency.The sorting efficiency is about 40 pcs/min.