摘要
为了对鸡种蛋孵化早期胚胎进行性别鉴定,构建了高光谱图像采集系统,在400~1 000 nm范围内获取94枚种蛋孵化0~12 d的高光谱透射图像。分别在胚胎的圆头、中间、尖头3个部位选择感兴趣区域（region of interest,ROI）,获取400~1 000 nm波段的响应信号,构建了支持向量机（support vector machine,SVM）、偏最小二乘判别分析（partial least squares discriminant analysis,PLSDA）和人工神经网络（artificial neural network,ANN）的鸡胚胎性别鉴定模型,并比较了不同孵化时间雌雄胚胎的鉴别准确率。试验结果表明,SVM模型、PLSDA模型和ANN模型均对孵化第10天种蛋中间部位检测效果最好。随后通过分析第10天种蛋中间部位光谱响应的差异,选取600~900 nm的光谱值构建胚胎性别鉴定模型,结果发现,3种模型的判别准确率均有上升,SVM模型和PLSDA模型预测集样本判别准确率均为75.00%,ANN模型预测集样本判别准确率达到82.86%。其中,ANN构建的种蛋孵化胚胎性别检测模型的整体效果优于SVM模型和PLSDA模型。结果表明高光谱图像技术在检测鸡种蛋孵化早期胚胎性别方面有一定效果,但种蛋蛋壳的个体差异会对鉴定准确率造成一定影响。
In poultry breeding, chicken gender is closely related to its productivity. For the rapid and nondestructive detection on the gender of chicken embryos in early period during the whole hatching, a laboratory hyperspectral imaging system was setup to capture hyperspectral transmission images of 94 hatching eggs on days 2, 4, 6, 8, 10 and 12, at the spectral region of 400 ~1 000 nm. On the 16 th day during the hatching, chicken egg embryos were broken for gender determination by human visual inspection from three trained experts, due to the fact that the difference in morphology between male and female chicken embryos is enough to be identified by human vision on day 16. After comparisons among the images information between male and female embryos, no clear differences was found. Therefore, the spectral information was further analyzed. The regions of interest（ROIs） were selected at big head, middle part and small head of egg for spectral response extraction. For the optimal time and location selection for gender determination, three models including linear algorithm（partial least squares discriminant analysis, PLSDA） and nonlinear algorithm（support vector machine, SVM; artificial neural network, ANN） were built by using full band spectral response, and the discrimination accuracy for male and female embryos were compared at different time and locations based on different models. The results showed that all of the discrimination models had the best accuracy on the 10 th day of incubation as well as the ROI located on the middle of egg. Next, the average spectral response difference between male and female embryos was analyzed. 600 ~900 nm was determined by removal of bands of 400~599 nm and 901~1 000 nm, which had little difference between male and female embryos, to build the gender discrimination models based on SVM, PLSDA and ANN. The results showed that the accuracy of SVM and PLSDA were both 75.00% for prediction, which were better than the accuracy with 71.88% for both SVM and PLSDA, based on full band spectral response on day 10 and middle part. For ANN, the prediction accuracy was 82.86%, by an increase of 2.86% compared with full band spectral response. Furthermore, the overall prediction performance of ANN model had better results for gender detection than SVM and PLSDA. The results in the study showed that hyperspectral imaging technology has the potential to identify the gender of early hatching chicken eggs embryos, but the accuracy of identification may be affected by the individual differences of hatching egg. So, in further research, more samples, physicochemical and biochemical differences between male and female embryos in various developmental stages were suggested to be incorporated for higher accuracy acquision of gender discrimination.
出处
《农业工程学报》
EI
CAS
CSCD
北大核心
2016年第1期181-186,共6页
Transactions of the Chinese Society of Agricultural Engineering
基金
国家科技支撑项目(2012BAD28B01
2015BAD19B03)
中央高校基本科研业务费专项(KYLH201504)
关键词
光谱分析
图像处理
无损检测
高光谱图像
鸡种蛋
孵化早期
胚胎
性别鉴定
spectrum analysis
imaging processing
nondestructive examination
hyperspectral image
hatching egg
early incubation
embryos
gender determination
