Research Of Egg Sorting System Based On Machine Vision

With the development of machine vision technology and machine learning technology,automatic detection and sorting tasks based on machine vision are increasingly used in agriculture.In order to solve the problem of low efficiency and low accuracy caused by the traditional manual sorting method of cracked eggs in the initial processing line of eggs,an egg classification method based on machine vision and an evaluation method of eggshell damage degree are proposed through comparative research.Fast and accurate sorting of eggs.First of all,according to the functional requirements of the egg sorting system,the overall scheme of the system is designed.The system is divided into:visual inspection module,transfer module,sorting module and communication module;and based on the existing hardware foundation of the egg preliminary processing production line.Preliminary selection and description of the hardware of each module are carried out,and the development environment of the software part is also explained.On this basis,it focused on the solution that can realize the online collection of egg shell images,selected the camera,lens model,and light source type,completed the collection and labeling of the data set,and designed a set of egg image preprocessing procedures.,The segmentation of the crack area in the data set image is realized.Secondly,in order to realize the classification of cracked eggs and intact eggs,an egg image classification method based on SVM and an image classification method based on improved CNN are proposed respectively,and a comparative study:After feature extraction of the image processing results,input SVM for training and Tuning the hyperparameters obtained an egg image classification model based on SVM with a comprehensive accuracy rate of 93.0%.An egg image classification algorithm based on improved CNN was proposed.This algoritlim is aimed at the problem of large calculation amount of ordinary convolution in CNN.The use of deep separable convolution instead of traditional convolution layer reduces the calculation amount and time consumption of a single prediction.After training and tuning,the comprehensive accuracy rate reaches 97.3%,which is better than the classification method based on SVM.Finally,according to the image acquisition plan and image classification model,the logical idea of realizing egg crack detection and sorting based on the image classification method is explained.Based on the classification results of multiple images,a comprehensive judgment is made on each egg,and the judgment result is extracted as a crack.The related information of the eggs is evaluated on the degree of damage to the eggs,and then the statistics are output in the GUI interface designed according to the requirements of the factory.