Research On Chicken Wings Quality And Mass Detection Based On Machine Vision

In recent years, following the production of chicken continues growing in China, the total output is in the first three of the world and chicken has been becoming the second meat consumer goods after the pork. Chicken wings are popular among consumers as an important poultry meat product. With the development of society, the quality of people’s life and consumption level has been improving gradually so that the demand of poultry meat product is higher and higher, people hope that poultry meat product is not only health but also fresh and delicious.Currently, classification of chicken’s quality or mass is used generally by manual work.The method takes time and energy, have a lower efficiency and easily influenced by subjective factors. To guarantee poultry classification processing quality and improve the level of poultry processing automation, this paper aims to design a method of detecting chicken wing’s quality and forecasting its mass based on machine vision technology. The main contents are as following:(1) Build machine vision system. System includes both hardware and software. The choice of industrial camera, gathering background will influence the result of the image processing. Collect chicken wing images and build image acquisition library, and optimize the acquisition environment.(2) Using the image processing technology to process the images with gray processing,image enhancement, morphology denoising processing and so on. Analysis of the features of chicken wings, extract the area percentage of congestion part and respectively extract the area,outline circumference, long and short axis from the top view and front view of wings’ binary image. And extract color features respectively from RGB, HSI and Lab color models.(3) Set up the quality detecting model and the mass forecasting model. Using the principal component analysis to analyze the area percentage of congestion part features and color features, and remove redundant information. Select the grid search method to optimize the model parameters, and establish the quality detecting model. Calculate the actual characteristic values of chicken wing shape through the image size calibration method. Using function fitting method to build forecasting models of one-dimensional linear, power exponent, multi-linear and multivariate mixed function respectively. Experiment results show that, the best forecasting model is multivariate mixed function, and the correlationcoefficient2 R is 0.972, estimation error variance2 S is 0.506. So select it as the optimal mass forecasting model.(4) Verify and analyze the two models. Using new data to test the quality detection model, and prediction accuracy reached 98.8%. Analysis of misjudgment samples to find out the reasons of misjudgement. Using the test data to validate the prediction model, get the mean absolute error of forecasted mass and the actual mass is 0.266 g, and the mean relative error is 0.57%. Tests have shown that the method of detecting chicken wings’ quality and forecasting mass based on machine vision technology is feasible, and it can be applied to chicken wings’ mass grading.