On-line Grading System Of Chicken Carcass Quality Based On Deep Camera And Machine Vision Technology

Chicken carcass quality classification is one of the most important processes in chicken carcass production,processing and pre-marketing.From the whole chicken production line,the product classification,packaging,price and carcass quality classification are closely related.At present,the quality grading of chicken carcass has not been fully automated.Many small and medium-sized companies mainly use manual weighing and grading,which results in low efficiency and frequent mistakes in grading.Although large-scale slaughter and processing enterprises use automatic equipment for grading,they use contact weighing method for quality grading,which results in frequent contact with weighing instruments and secondary contamination of chicken carcass,which greatly increases the safety risk of food.Considering the above situation,there is an urgent need for a non-contact classification method in the market,which not only ensures accurate classification,but also solves the problem of secondary pollution of chicken carcass.According to the non-contact characteristics of image processing,this paper studies the method of chicken carcass classification based on depth camera and machine vision technology.The three-dimensional information of chicken carcass was collected by depth camera,and the linear and non-linear mathematical models for prediction of chicken carcass quality grade were established based on three-dimensional characteristic parameters and three-dimensional plus two-dimensional characteristic parameters respectively.The main contents and conclusions were as follows:(1)Establishing a linear mathematical model of quality grade prediction based on three-dimensional characteristic parameters of chicken carcass,and conducting two experiments.The first experiment was based on the depth information of chicken carcass breast.Using Kinect 2.0 depth camera,only the chicken carcass on the breast was photographed.Assuming that the volume of chicken carcass breast and back was the same,the volume of chicken carcass on the breast was calculated by using the mathematical method of triple integral.The volume of whole chicken carcass was multiplied by two times,and the linear mathematical model between the volume and mass of chicken carcass was established.The fitting degree of quality prediction is 0.952,and the classification accuracy is 91.4%.In the second experiment,based on the depth information of the breast and back of chicken carcass,Kinect 2.0 depth camera was used to photograph the breast and back of chicken carcass respectively.Similarly,the volume of breast and back of chicken carcass was calculated by triple integral method.The sum of the positive and negative volumes was the volume of the whole chicken carcass,which was also used to establish the relationship between the volume and mass of chicken carcass.The linear mathematical model shows that the fitting degree of the quality prediction reaches 0.9983,and the classification accuracy reaches 95.2%.The fitting degree of the quality prediction model of test 2 is slightly better than that of test 1.(2)Establishing a non-linear mathematical model of quality grade prediction based on three-dimensional and two-dimensional characteristic parameters.Seven feature quantities describing the size of chicken carcass were extracted from the pre-processed depth images.The two-dimensional feature parameters were projection area Sp,carcass length Hp,contour length Cp,chicken breast length Ap,chicken breast width Bp,chicken breast area Ep,and the three-dimensional feature parameter was chicken breast depth value Hd.Three machine learning methods,Random Forest(RF),Gradient Boosting(GB)and Gradient Boosting(AB),were used to establish a non-linear regression model for the prediction of chicken carcass quality grade.The R2 value of Gradient Boosting(GB)was 0.9960,and the RMSE value was 0.0390.The quality prediction model was obviously superior to the other two.In other models,the success rate of quality classification using the Gradient Boosting(GB)model is 95%.(3)A chicken carcass quality grading platform was designed.The quality grading system consists of image acquisition part,image processing part and grading execution part.Kinect 2.0 depth camera is responsible for image acquisition,industrial computer is responsible for image preprocessing and judgment of chicken carcass grade,Siemens S7-200 PLC is responsible for the control of pneumatic devices,complete the online quality grading work automatically.The quality classification experiment is carried out on the designed grading platform,and the results are analyzed and evaluated.Finally,through two groups of experiments,the classification results of the platform at different speed of the transport chain were tested.The results showed that the classification accuracy of the platform was 87.5%.