生菜叶片绿度的高光谱判别方法研究

Study on the Hyperspectral Discrimination Method of Lettuce Leaf Greenness

生菜叶片绿度在作物生理及品质感官评价中具有重要作用。结合目前高光谱检测与分析技术在植物生理信息监测中的应用现状,开展了基于高光谱技术的生菜叶片绿度判别方法研究,以此为叶菜品质感官评价的定量化及基于高光谱技术的多功能生理信息同步采集装置的开发提供必要的理论支撑。本文以生菜为研究对象,在三种不同光照强度下开展栽培试验。以叶绿素相对含量(SPAD)作为反应绿度的参数,获取生菜整个生命周期中的动态高光谱和SPAD数据,分析了高光谱曲线的变化规律,建立了高光谱与SPAD之间的关系模型。采用Savitzky-Golay卷积平滑(SG)方法对原始高光谱数据进行降噪,平滑后的数据分别与多元散射校正(MSC),标准正态变量变换(SNV)和一阶导数(FD)三种预处理方法组合,采用竞争性自适应重加权取样法(CARS)和提取有效植被指数(VI)两种方法进行敏感波长提取,结合偏最小二乘(PLS)和最小二乘支持向量机(LSSVM)两种方法建模,以决定系数(R^(2))和均方根误差(RMSE)为评价指标,优选出最优绿度判定模型。结果表明:在10,20和30d的生菜全生命周期内,不同光照强度下的高光谱曲线表现出总体变化趋势一致但反射率值不同的特征,在可见光450~680nm范围内,自然光照条件下的生菜高光谱反射率值要高于补光处理条件下的反射率值;而在近红外730~850nm范围内,生菜叶片的高光谱响应特征恰好与可见光范围内相反。基于SG+FD预处理与CARS敏感波长提取方法的组合可实现叶绿素相对含量特征信息的最有效提取,提取的敏感波长占全波长的64.59%,与原始高光谱(1.25%)相比,提取的敏感波长数增加了63.34%。最终确定LSSVM方法为最优建模方法,基于SG+FD+CARS+LSSVM组合方法所建模型为最优生菜绿度判定模型,训练集R_(e)^(2)=0.9207,RMSEC=1.1610,预测集R_(p)^(2)=0.8288,RMSEP=2.4008,模型精度较高,可以实现生菜叶片绿度判别的目的。

Lettuce leaf greenness is important in the physiological and sensory evaluation of crop quality.Based on the comparison of existing methods for greenness discrimination,combined with the application status and prospects of hyperspectral detection and analysis technology in the detection of plant physiological information,the research on the application method of hyperspectral technology in the greenness discrimination of lettuce leaves was carried out.The quantification of sensory evaluation of the vegetable quality and developing a multifunctional synchronous collection device for physiological information based on hyperspectral technology provide necessary theoretical support.Lettuce is the subject of study.Cultivation experiments were conducted under three different light environments,and relative chlorophyll content(SPAD)was used as a parameter to respond to greenness.Acquisition of dynamic hyperspectral and SPAD data throughout the life cycle of lettuce.Study of hyperspectral response characteristics to leaf greenness.The variation pattern of the hyperspectral curve was analyzed.Finally,a relationship model between hyperspectrum and SPAD was developed.The Savitzky-Golay convolution smoothing(SG)method was used to reduce the noise of the original hyperspectral data.The smoothed data was combined with the three preprocessing methods of multivariate scattering correction(MSC),standard normal variable transformation(SNV)and first derivative(FD),and finally adopted competitive adaptive reweighted sampling(CARS)and extraction effective vegetation index(VI)two methods for sensitive wavelength extraction.Combine the two methods of partial least squares(PLS)and least squares support vector machine(LSSVM)for modeling,and use the coefficient of determination(R^(2))and root mean square error(RMSE)as evaluation indicators to select the optimal greenness prediction model.The results showed that the hyperspectral curves of lettuce under different light environments showed a consistent overall trend but different reflectance values during the whole life cycle of lettuce at 10,20and 30days.The lettuce reflectance values in the visible light range of 450~680nm exhibited higher natural light exposure than the supplemental light treatment,while the hyperspectral response characteristics in the NIR range of 730~850nm were exactly opposite to the visible light range.The combination of SG+FD pre-treatment and CARS sensitive wavelength extraction method based on SG+FD can achieve the most effective extraction of chlorophyll content feature information,and the extracted sensitive wavelengths accounted for 64.59%of the total wavelengths,which increased the number of extracted sensitive wavelengths by 63.34%compared with the original hyperspectrum(1.25%).The LSSVM method was identified as the optimal modeling method,and the model built based on the combined SG+FD+CARS+LSSVM method was the optimal lettuce greenness prediction model with the training set R_(e)^(2)=0.9207,RMSEC=1.1610,and the prediction set R_(p)^(2)=0.8288,RMSEP=2.4008,indicating that the model had high accuracy.The purpose of greenness judgment of lettuce leaves can be realized.