利用高光谱透射图像检测鸡种蛋早期孵化  

Detecting early embryo development of chicken hatching eggs by hyperspectral transmittance imaging

作　　者：张伟[1] 潘磊庆[1] 屠康[1] 

机构地区：[1]南京农业大学食品科技学院,南京210095

出　　处：《农业工程学报》2012年第21期149-155,共7页Transactions of the Chinese Society of Agricultural Engineering

基　　金：国家自然科学基金(31101282);中央高校基本科研业务费专项资金(KYZ201120);农业部948项目(2011-S10);江苏省普通高校研究生科研创新计划资助(CXZZ12_0275)

摘　　要：为了检测鸡种蛋孵化前期胚胎发育情况,构建了高光谱图像采集系统,在400～1000nm范围内获取90枚种蛋孵化前3天的高光谱透射图像。通过独立分量方法对高光谱数据进行分析降维,优选出571、614、661、691和716nm共5个特征波长,提取每个波长下的光谱平均值和标准偏差,获得每个样品10个特征变量。为了消除变量之间相关性,利用主成分分析提取了4个主成分变量,在此基础上构建了学习向量量化(LVQ)神经网络判别模型。验证性试验均表明该模型具有较高的稳定度(变异系数为1.7),对第1,2,3天的测试样本判别准确率分别为78.8%,90.3%和98.6%。结果表明高光谱图像技术可以检测种蛋孵化前期胚胎发育情况。In order to detect chicken hatching egg incubation during the early period, a laboratory hyperspectral imaging system was setup to capture hyperspectral transmission images of 90 hatching eggs on the first three days at the spectral region of 400～1 000 nm. Dimension reduction was implemented on hyperspectral data based on Independent Components Analysis （ICA） and 5 characteristic bands with 571、614、661、691 and 716 nm. Next, spectral average and standard deviation were extracted from each band, thus 10 characteristic variables in total for 5 characteristic bands were acquired. To remove the correlation between variables, Principal Component Analysis （PCA） was conducted on 10 characteristic variables, and 4 principal component variables were extracted as the input of the discrimination model constructed by Learning Vector Quantization （LVQ） artificial neural network. Verification experiments showed that discrimination model had good stability （cv 1.7） and achieved prediction accuracies of 78.8% on 1 st day, 90.3% on 2 nd day and 98.6% on 3 rd day. This research demonstrates that the hyperspectral imaging technique is feasible for detecting hatching eggs incubation during the early hatching period.

关 键 词：光谱测定法 独立分量分析 神经网络 检测 孵化 种蛋 

分 类 号：S126[农业科学—农业基础科学] TP391.4[自动化与计算机技术—计算机应用技术]