基于NIR的主成分结合支持向量机鉴别蚕茧雌雄的研究

Study on the method of determining cocoon sexuality by PCA plus SVM based on NIR

研究应用近红外光谱分析技术鉴别蚕茧雌雄的方法。削茧取蛹,辨别蚕蛹的性别,将雄茧设为1,雌茧定设为0。用自制附件获得蚕茧的近红外漫反射光谱(11 300～3 500)cm^(-1),其中校正集20枚,预测集19枚。对光谱做软阈值小波消噪、5点一阶微分及正规化处理。提取主成分,计算累积贡献率,给前2个及3个主成分作图。分别取1～10个主成分并结合不同的ε(10^(-1)～10^(-9))用支持向量机建立预测模型。结果表明,从光谱的前3个主成分图中不能区分蚕茧的雌雄,主成分相对分散,前10个主成分累积贡献率才达85.23%。所建模型的预测正确率随着所取主成分个数增加而升高,随着ε取值的下降而先升后降。当主成分数为6～10、ε为10^(-5)时,预测正确率最高,达94.7%。本研究为利用近红外光谱鉴别蚕茧雌雄提供理论依据。

Study on the method of determining Cocoon sex by NIR. 20 samples from total 39 Cocoon was selected randomly for calibration set, the other 19 Cocoon was used as prediction set. Cocoon sexuality was identified through cutting the cocoon and examining its pupa, the male and female sexuality was set to values of 1 and 0, respectively. After diffuse reflection, spectrum t （ 11 300 - 3 500）cm^-1 was get, it was denoised by wavelet arithmetic, and was processed by 5 point 1 rank differential coefficient and normalization. The principal component of spectrum was picked up, accumulating contribution of principle component was acquired, and distribution figure of the 2 and 3 principle component was draw. It was established that the model of SVM with 1 to 10 principal components and 9 different ε（ 10^-1 to 10^-9 ）. The results showed that the accumulating contribution of 10 principle components was 85.23% and the Cocoon sexuality can not be determined through distribution figure of 2 or 3 principle component. The right rate of prediction was rise with the number of principal components used, and it rise in the beginning and down afterward with decline of ε. The best rate of prediction is 94.7% in the condition of 8 equaled to 10^-5 and 6 to 10 principal components was adopted. This study will provide a novel way for determining Cocoon sexuality in practice.