Research On Key Technology Of Automatic Fruit Sorting System Based On Machine Vision

With the rapid development of machine vision technology,automatic sorting technology based on machine vision has been applied to various production and processing fields.In order to solve the problem of low efficiency and high error rate of traditional sorting methods in the case of huge fruit yield,a new method for accurate detection of fruit position and rapid classification of apple species was proposed by using machine vision and deep learning techniques.It can effectively improve sorting efficiency and accuracy.This paper mainly studies these two key technologies in the fruit sorting system and achieves high sorting accuracy.First,the overall scheme of the automatic sorting system was designed according to the functions and requirements required by the automatic sorting system.The system is mainly divided into hardware design part and software design part.The hardware part is mainly composed of image acquisition module,robot module and conveyor module.The image acquisition module is used for image acquisition.As a data source for image analysis,the robot is used to capture the fruit.The software part is based on the typical apple in the fruit.It is mainly divided into two parts:apple position detection and classification identification.Secondly,the method of position detection of apple is studied.The camera is calibrated by Zhang Zhengyou calibration method to determine the relationship between the three-dimensional geometric position of the surface of the space object and its corresponding point in the image.The operator performs edge detection,and then gives a method to extract the key points of the center by edge contour detection.It can be converted into a robot coordinate system by coordinate transformation to perform subsequent operations such as grabbing.Finally,the classification and recognition technology of apple is studied.This paper studies six common apples as research objects,and proposes a deep learning method based on convolutional neural network to obtain several apple image data from different data sources.Labeling,using 80*80*3 pixel color image as model input,adding batch normalization layer,changing the over-fitting strategy,finding the optimal parameters in the model through contrast experiments,and finally adjusting the model and parameters.In the case of excellent results,a classification accuracy rate of 92%was obtained.And the model has a certain degree of extensibility.By adding different fruit data sources for model training,a variety of fruits can be identified.