Research On Intelligent Dynamic Preferential Equipment For Scaled-up Winter Jujubes Based On Convolutional Neural Network

As a unique fresh fruit in China,winter jujube has become an economic pillar industry in many regions.However,the current winter jujube production and subsequent processing rely heavily on manual work.Especially winter jujube screening,which is not only time-consuming and labor-intensive but also has many drawbacks.And there is an urgent need for an equipment that can replace manual sorting.Fruit sorting is not only an important step before it enters the market or storage,but also an important method for quality assurance and improving market competitiveness.Therefore,in order to improve the sorting efficiency and accuracy of winter jujubes and reduce the human and material investment,we designed a device for detecting the surface quality of winter jujubes based on convolutional neural network technology and computer vision technology in this paper.Aiming to promote the development of the winter jujube industry.This study was based on the actual production status of Zhanhua winter jujube.The main contents are as follows:(1)Structural design of dynamic winter jujubes sorting equipment.We investigated the current production situation of the jujube industry in the Zhanhua area.Analyzed the structure,working principles and characteristics of existing equipment.Designing automatic testing equipment and systems for winter jujubes quality according to date sorting requirements.Perform the necessary theoretical calculations and simulation analysis for each structure.Finally,the theoretical model of the date inspection equipment was established using 3D software.(2)The design of the reject control system and the drive system of the device.Both of them are controlled by an Arduino UNO development board.The rejection control system converts the received detection results into binary.The 74HC595 chip realizes the simultaneous pushing of multi-channel signals.The optocoupler relay switch receives the signal controls the solenoid conduction and closure to realize the shunting of different quality winter jujubes.The driving system adopts 86HS120 type stepper motor.The pulse signal is provided by Arduino UNO.And the controller is MA860 H.(3)In this paper,we first establish the datasets for two schemes of dichotomous classification and detailed classification for the surface quality of winter jujube based on the grading standard of winter jujube and the actual production demand.Then,the traditional improved Alex Net model.Training experiments are conducted with several different neural network to compare and analyze the training effects of each model.It was verified that among the two classification models.The improved Alex Net model achieved98% accuracy on the validation data,the Inception V3 model achieved 97% accuracy,and the Inception V3 model for detailed classification achieved 95% accuracy.(4)Proposes to use machine vision technology to pre-process the captured images.Combining the characteristics of date images and HSV color space.Red and orange are used as the recognition interval for target capture.The obtained images are processed by erosion,expansion,and filtering,etc.Then the target in the image is extracted and segmented.In the end,the results are fed into the network and derived by size transformation.In addition,to ensure the comprehensive detection of winter jujube,adopting the superposition algorithm.Once the detection results of captured winter jujube are defective.The final returned instructions are all rejection.Finally,the visual inspection system was verified and proved that the research in this paper can realize the work of date surface inspection.In terms of efficiency,it can process about 29,000 winter jujubes per hour,and their weigh about 530 kg.