Detection Of Residual Eggshells On The Surface Of Salted Egg Yolk Based On Machine Vision

China is a major producer of poultry eggs,ranking first in the world in terms of total egg production and consumption.There are many kinds of poultry and egg foods.As one of the traditional Chinese foods,salted egg yolk has always had a good reputation and market.At the same time,it is widely used as the raw material of rice dumplings,moon cakes and other foods.At present,the automation level of salted egg yolk processing equipment is not high,most processes rely on labor,the production line efficiency is low,and the production cost is high.At the same time,manual operation also has serious food safety hazards.Therefore,the research and development of salted egg yolk automatic production equipment,especially the visual inspection technology and equipment for salted egg yolk quality defects,is a practical technology urgently needed by salted egg yolk processing enterprises.This paper focuses on the online detection of residual eggshells in salted egg yolks,and carries out corresponding research and obtains the following conclusions:(1)The overall scheme,structure and working principle of the salted egg yolk automatic production line were analyzed;the machine vision detection method was established through the comparison of various detection methods;the image acquisition platform was built,and the human-computer interaction interface of the detection system was developed.(2)Proposed a tracking detection algorithm based on YoloV3 model and DeepSort algorithm.Used YoloV3 model calculate the product prediction frame,the DeepSort algorithm was used to track the obtained prediction frame,the image features of the product are extracted for product classification and marking,and the residual eggshell was identified.Used the Opencv to realize the functions of counting,positioning and saving pictures of classified products.The experimental results showed that this solution solved the problem of tracking and detection of sports products.The detection accuracy of residual eggshells was 0.99,the recall was 0.98,the F1 value was 0.99,and the Iou reached 91.40%.(3)A detection algorithm for salted egg yolk products based on convolutional neural network was proposed.Four mainstream classification networks,VGG19,Resnet50,InceptionV3,and Inception ResnetV2 were compared,which were used to classify salted egg yolk products into two categories: no eggshell residues and those with eggshell residues,and the VGG19 model was the best.In order to improve the model efficiency and reduce the amount of computation,the VGG19 model is optimized.The optimized VGG19 model detects the residual eggshell of salted yolk: ACC=0.9745,PPV=0.9693,TPR=0.98,TNR=0.969,F1=0.9746.The detection accuracy has been further improved.(4)A residual eggshell detection algorithm combining convolutional neural network and random forest algorithm was proposed.The optimized VGG19 model and Resnet50 model were used to extract the features of the residual eggshells of salted yolk,and then the features were sent to the random forest classifier,and the classifier was trained to calculate the appropriate node weights to obtain high-precision residual eggshell identification results.The residual eggshell area in the salted egg yolk can be extracted separately,and the size of the residual eggshell area can be calculated and classified into4 categories: when the area ratio is less than 0.5%,qualified product,and when the area ratio is between 0.5% and 1%,partly defective product.When the area ratio is between 1%and 2%,defective product,and when the area ratio is higher than 2%,it is a waste product.Experiments show that the combination of convolutional neural network and random forest algorithm makes the detection scheme have good anti-interference,generalization and high efficiency.(5)After synthesizing the development and experiments of the above detection algorithms,this paper aims to identify the residual eggshells in the enterprise salted egg yolk and the production requirements and accuracy indicators of the salted egg yolk product classification,and determine the final detection scheme of the salted egg yolk product.Firstly,the YoloV3 model and DeepSort algorithm are used to realize the tracking and detection of salted egg yolk products,and at the same time,the identification of salted egg yolk with residual egg shell is completed,and then the image of the salted egg yolk with residual egg shell(including the salted egg yolk detected by mistake due to surface reflection)is saved.Input to the optimized VGG19 model to detect again,identify the misdetected salted egg yolk,and increase the detection accuracy.The convolutional neural network and random forest algorithm were used to extract the area of the residual eggshell area,and the proportion of the residual eggshell area was calculated for classification.The comprehensive detection scheme of residual eggshell in salted egg yolk proposed in this paper,the established model and the developed software can meet the technical requirements of enterprises for automatic identification and classification of residual eggshell in salted egg yolk,and provide technical support for the automation of salted egg yolk product production.