受精种蛋理化指标与光谱信息的响应关系

The response of physical and chemical indicators of fertilizedbreeder eggs based on spectral information

[目的]基于高光谱技术探究受精种蛋理化指标与光谱信息的响应关系。[方法]以海兰灰种鸡蛋为研究对象,通过测定孵化0~4 d受精与未受精种蛋蛋白质、脂肪、水分、pH值和颜色参数的变化,同时采集高光谱透射图像,确定受精与未受精种蛋存在显著差异的理化指标;将受精种蛋理化指标与光谱波段的透射率进行相关性分析,筛选特征波段;建立全波段(400~1 000 nm)范围种蛋受精的PLS-DA(偏最小二乘判别分析,partial least squares-discriminant analysis)和SVM(支持向量机,support vector machine)判别模型以及特征波段范围的受精判别模型,并对结果进行对比。[结果]在孵化0~4 d,受精与未受精种蛋蛋白质、脂肪、pH值和a~*值存在显著性差异,其中蛋白质、脂肪和a~*值与光谱波段的透射率呈负相关关系;pH值与光谱波段的透射率呈正相关关系;通过相关性分析,分别筛选出20、40、60和80个特征波段。基于全波段的建模预测集准确率在4 d达到最高,为100.00%。基于特征波段的建模预测集准确率在特征波段为60和80时达到最高,均为97.50%。[结论]光谱信息可以很好反映受精种蛋的理化指标变化,为高光谱技术检测种蛋受精提供了一定的理论基础。

[Objectives]The response of physical and chemical indicators of fertilized breeder eggs based on spectral information was investigated.[Methods]Using Hyland Gray breeder eggs as the research object,the changes of protein,fat,moisture,pH value and color parameters between fertilized and unfertilized eggs at 0-4 d of incubation were measured and hyperspectral images was obtained,the physicochemical indicators that were significantly different between fertilized and unfertilized eggs were determined;the physicochemical indicators of fertilized eggs were correlated with the transmittance of spectral bands and the characteristic bands were screened.PLS-DA(partial least squares-discriminant analysis)and SVM(support vector machine)discriminant models for egg fertilization in the full wavelength range(400-1000 nm)and discriminant models for fertilization in the characteristic wavelength range were established.[Results]At 0-4 d of incubation,there were significant differences between fertilized and unfertilized eggs in terms of protein,fat,pH and a value,of which protein,fat and determined a value were negatively correlated with the transmittance of spectral bands;pH value was positively correlated with the transmittance of spectral bands.Feature bands of 20,40,60 and 80 were screened out based on correlation analysis respectively.The accuracy of the modeling prediction set based on the full band range reached the highest at 4 d with 100.00%,and the accuracy of the modeling prediction set based on the feature band reached the highest at the feature bands of 60 and 80 with 97.50%.[Conclusions]Spectral information could well reflect the changes of physicochemical indexes of fertilized breeder eggs,which provided a tool to detect egg fertilization with hyperspectral technology.