基于可见/近红外光谱技术的黄瓜叶片SPAD值检测

DETECTION OF SPAD VALUE OF CUCUMBER LEAVES BASED ON VISIBLE/NEAR INFRARED SPECTROSCOPY TECHNIQUE

为了快速准确检测黄瓜叶片的SPAD值,采用可见/近红外光谱技术并结合化学计量学方法建立了黄瓜叶片SPAD值校正模型.并用不同建模方法对全波段光谱进行建模,结果表明用最小二乘支持向量机（LSSVM）建模得到的预测效果最好,其相关系数r和预测均方根误差RMSEP分别为0.9583和0.9732.通过分析黄瓜叶片的光谱反射率与SPAD值的相关系数和PLS建模回归系数,得到了531-581nm和696-716nm 2个特征波段以及556nm、581nm、698nm和715nm 4个特征波长,应用LSSVM分别对特征波段和特征波长建模.分析表明,采用特征波段建模,其预测相关系数r和预测均方根误差分别为0.9338和1.1370,与全波段建模结果相近,而采用特征波长建模效果稍差.特征波段建模大大减少了建模中的运算量,提高了建模速度,便于相应检测仪器的开发,所以,采用光谱特征波段建模对黄瓜叶片SPAD值的检测更为有效.

For the rapid detection of SPAD value of cucumber leaves, the calibration model of SPAD value was built by using visible and near infrared （Vis/NIR） spectroscopy technique and chemometrics methods. Different calibration methods were used to build the model in the whole wavelength region. The results indicate that the optimal performance is achieved by least squares support vector machine （LSSVM） model, and the correlation coefficient （r） and root mean squares error of prediction （RMSEP） are 0. 9583 and 0.9732, respectively. Via the analysis of correlation coefficients between the spectral reflectance and SPAD values, and the regression coefficients of partial least squares （PLS）, two characteristic wavelength bands （531 - 581nm and 696 -716nm） and four characteristic wavelengths （556, 581, 698 and 715nm） were obtained. LSSVM was used to the aforementioned wavelength bands and wavelengths. The results indicate that the characteristic wave- length bands can achieve a better performance with r of 0.9338 and RMSEP of 1. 1370. The prediction results are similar to the whole wavelength region model, while, the performance of LSSVM model with four characteristic wavelengths was not satisfying. Calibration method of using characteristic wavelength bands can largely reduce the calibration computation, and increase the calibration speed. Hence, it is more effective to use characteristic wavelength bands for the detection of SPAD values of cucumber leaves.