Design And Research Of Blood Orange Sorting Machine Based On Deep Learning

As one of the important origin of citrus,the planting scale of citrus industry has continued to grow in recent years in our country.The total output of citrus has reached5595.61 million tons in 2021.Citrus grading can not only improve its commodity value,but also help to build a brand to improve the visibility of high-quality products and increase their market competitiveness.The current design of citrus sorters have defects such as large size,high price or single classification standard.It is difficult for small and medium-sized farmers to purchase large intelligent sorting machines,small citrus sorting machine with a single classification standard,insufficient degree of insufficient,For these reasons,this paper designed and developed an intelligent small citrus sorter is of great importance.In this paper,Taroko blood orange as the research object,combined with citrus grading standards and the design requirements of fruit grading machine.And using experimental optimization,sensors,automatic control and machine vision technology to design and develop an intelligent blood orange sorting machine for solving above problems.The main research content is as follows:(1)Design and test of the overall plan of the whole machine and its key structuresThe design of the whole machine is determined by analyzing the blood orange grading requirements,and key structures such as baffle feeding module,black box identification module and disc sorting tray module are designed,combining the fruit shape characteristics and fruit diameter size of blood orange,motor speed,lifting inclination angle,brush turning angle and baffle height are used as test factors.Qualification rate(Q)as the test index.Single-factor test and orthogonal test on the feeding module of the whole machine.Determine the range of variation of each factor value and the optimal combination based on the analysis of experimental results,and CCD tests are conducted on the above 4 factors,and use Design-expert13 software to analyze the surface response of the test results and optimize the parameters,the test results show that the factors affecting the qualification rate(Q)are in the following order:motor speed>lifting inclination>baffle height>brush angle.The optimal combination of parameters is 55r/min motor speed,35＞lifting angle,26.9255mm baffle height and 37.0254 brush angle,the pass rate reaches the maximum 95.7071%,best conveying performance at this time.(2)Design of control systemAccording to the functions that the sorter needs to achieve,this paper uses PC and STM32 as the controller to build the whole hardware control system.The system mainly contains STM32 control module and computer vision control module;feeding module,position detection module as well as image acquisition module,sorting module.The developed control system can obtain the photoelectric sensor signal in current time,control the camera to obtain and classify blood oranges,drive the servo to rotate to complete the target action,and achieve the purpose of blood orange classification.In this paper,the software system uses C and Python as the system development language respectively,Based on the analysis of the hardware of the control system,the software program of the control system of the whole machine is designed according to the actual requirements.Software design and development for DC motor drive,position detection,image acquisition and classification subroutines,and sorting execution subroutines.Through the entire system software design,the hardware and software of the sorter control system cooperate with each other to make decisions and execute corresponding actions of the blood orange grading system.(3)Design of image classification systemThis paper builds an image classification system based on the Paddlepaddle framework.According to the established classification rules for blood oranges,the self-developed image acquisition platform is used to create blood orange datasets according to the image classification dataset format in paddle.And the dataset is divided according to training set,validation set and test set,each accounting for 70%,15%and15%.Res Net,Dense Net,Mobile Net and PP-LCNet are trained separately using the established datasets,based on the training results it is finally concluded that the Dense Net and Mobile Net models have higher convergence of Loss values than Res Net and PP-LCNet,although the Loss convergence values of Res Net and PP-LCNet are very close,PP-LCNet has the highest accuracy in the test set.The final prediction accuracy and summation mean of PP-LCNet in the validation set are 98.79%,which is the best performance and therefore identified as the final network for blood orange classification.In this paper uses paddlex to convert the training model into inference format for export,write the visual control code program through pycharm,and deploy it to the PC using the API corresponding to paddlex.(4)Classification performance test of the whole machineOn the basis of completing the prototype of blood orange sorting machine,referring to NY/T 2617-2014"Technical Specification for Quality Evaluation of Fruit Grading Machine",the experimental test was conducted on single type of fruits and mixed fruits of sorting machine respectively.The results of the test show that the sorting accuracy of this blood orange sorter for three categories of blood oranges was in the order of C1>A1>B1,and the corresponding probabilities are 99.00%,96.00%and94.33%.The average passing rate of three categories of blood oranges is 96.44%>95.00%,which meets the grader standard.The average accuracy of the 5 groups of mixed fruits is 95.47%,which is lower than the average accuracy of 96.44%for single fruit classification,but still greater than 95.00%,meeting the relevant standards in the"Technical Specification for Fruit Classifier Quality Evaluation".This study provides an effective solution to solve the problems of large labor intensity of blood orange grading,low intelligence of grading machinery,high cost of equipment and single grading standard,and provides a theoretical and technical basis for the research of small intelligent blood orange sorting machinery.