| In the traditional poultry industry,the breeding direction of chicken is egg and meat compatibility.The market under the new agricultural system put forward new requirements for the breeding of chicken,which requires the separation of eggs and meat to achieve a dedicated supply of egg and meat products.The detection of sex in chicken hatching eggs is the basis for achieving a dedicated supply of eggs and meat products.At present,the main sex detection of chicken hatching eggs in China is the artificial screening of young chicks after being out.This method is slow to detect and is easily influenced by human subjective factors.Therefore,in order to ensure that the market of two products just needs to be under the premise also avoiding slaughter behavior of young chicks after being out,the research of early detection of sex in chicken hatching eggs not only has ethical and moral significance,but also has practical economic value.In this study,the white shell of chicken hatching eggs was used as the object.Based on the image technology,a sex detection model for chicken hatching eggs was established to achieve sex detection in early hatching of chicken hatching eggs.The main research contents and conclusions of this topic are as follows:(1)Build the machine vision image acquisition system of chicken hatching eggs.The image acquisition system of this paper is mainly composed of industrial camera(Basler,Aca1600-20uc),lighting system and computer.Through the comparison of different light sources,it was determined that the light source for this study was a 1W LED lamp,which is installed at the bottom of illumination system.The eggs to be detected are placed laterally on the loading platform,transmitted through the light source under the loading platform,using the industrial camera to get the image and transfer to the computer through the USB data line.Comparing different acquisition methods,and determine the horizontal egg collection horizontally after 2 min using the LED lamp;the object distance of the image acquisition system is 218 mm.(2)Highlight the blood line texture in the hatching eggs image.The color images of chicken hatching eggs were collected at different hatching periods,also analyzed and compared with the images of 3d,4d,5d,6d,8d and 10 d.According to the clarity of the blood line and the principle of blood line integrity within the vision field of the machine vision,the image of the 4d of the early hatching was finally determined as the research object.The ROI region of the hatching eggs was determined by preprocessing the hatching eggs image to remove the image background color and salt noise.Using adaptive histogram equalization and combined morphological processing to enhance the contrast of the bloodline in the image,and the blood line texture is highlighted from the entire background.(3)Extract the characteristic parameters of chicken hatching eggs.The F test was used to analyze the significance of the egg weight loss rate and sex during the whole incubation period of the hatching eggs,and to exclude egg weight as a characteristic to detect the sex of the chicken hatching eggs.For the sex characteristics of eggs,data were quantified using the corresponding indicators.Using the minimum circumscribed rectangle,gray level co-occurrence matrix,histogram statistics and differential box dimension method respectively to extract 11 dimensional characteristics,which is egg-shaped index,energy,contrast,correlation,entropy,mean,standard deviation,smoothness,third order moment,consistency and fractal dimension.(4)Establish a sex test model of chicken hatching eggs.The first batch and the second batch of samples were randomly selected according to the principle of 2:1allocation of training set and prediction set consisting of sample 1 and sample 2;the first batch of samples was used as the training set,and the second batch of samples was used as the prediction set consisting of sample 3.The GA-SVM model was established for the early hatching chicken hatching eggs,and detection accuracy of the three groups of samples was 70.97%,63.33%,and 60% respectively;The GA-single hidden layer BP model was established for the early hatching chicken eggs,and detection accuracy of the three groups of samples was 80.65%,81.67%,and 80.56%,respectively;The GA-double hidden layer BP model was established for the early hatching chicken eggs,and the three groups of samples,and detection accuracy is 90.32%,90%,and 90%,respectively.The experimental results show that the GA-BP model based on the 11 dimensional hatching eggs image feature is superior to the GA-SVM model,which shows that it is feasible to use the machine vision image technology to detect chicken hatching eggs sex of early hatching;The accuracy rate of GA-double hidden layer BP model for hatching eggs sex detection in early hatching was higher than that of GA-single hidden layer BP model,and between groups and groups within the sample difference between detection accuracy is not much different,indicating that model has better robustness.The average accuracy rate of the three batches of samples can be used to express the comprehensive detection accuracy of the model,the comprehensive detection accuracy rate is 90.11%,and the model detection time is short.And average detection time of a single sample is0.2679 ms,which is to achieve the early hatching chicken hatching eggs sex online test laid the foundation. |
