高光谱成像的红提总酸与硬度的预测及其分布可视化

利用高光谱成像技术实现对红提总酸和硬度无损检测和分布可视化。首先，利用高光谱采集生长期360个红提样本在波段450～1 000 nm的高光谱图像信息后用化学方法测定对应样本的总酸，用质构仪测定硬度。采用KS(Kennard-Stone)算法将总样本按照3∶1的比例划分为训练集（270个样本）和测试集（90个样本）。对红提原始光谱数据分别利用标准正态变量变换（standard normal variate transformation,SNV）、卷积平滑（Savitzky-Golay,SG）处理法、多元散射校正（multivariate scatter correction,MSC）、归一化等光谱预处理方法处理，确定最优光谱预处理方法。然后，分别采用一次降维（竞争性自适应重加权算法（competitive adaptive reweighted sampling,CARS）、连续投影算法（successive projections algorithm,SPA）、遗传算法（genetic algorithm,GA）、无信息变量消除法（uninformative variable elim...

Hyperspectral Imaging for Prediction and Distribution Visualization of Total Acidity and Hardness of Red Globe Grapes

In this paper, hyperspectral imaging technology was used for nondestructive detection and distribution visualization of total acidity and firmness of red globe grapes. The hyperspectral information of 360 samples of growing red globe grapes in the wavelength range from 450 to 1 000 nm was collected using a hyperspectral instrument, and the total acidity and firmness of these samples were determined by titration and a texture analyzer, respectively. The Kennard-Stone (KS) algorithm was used to divide the total samples into a training set (270 samples) and a test set (90 samples) in a 3:1 ratio. The collected raw spectral data were preprocessed using various methods such as standard normal variate (SNV), Savitzky-Golay (SG), multivariate scatter correction (MSC), and normalization to determine the best spectral preprocessing method. Then, the feature variables were extracted from the spectral information using six dimensionality reduction algorithms: competitive adaptive reweighted sampling (CARS), successive projections algorithm (SPA), genetic algorithm (GA), uninformative variable elimination (UVE), CARS-SPA, and UVE-SPA. Using partial least squares regression (PLSR), optimal prediction models for total acidity and firmness were developed separately. Finally, the total acidity and hardness for each pixel of the hyperspectral image were calculated according to the proposed optimal prediction models, and a gray scale image was obtained and pseudo-color transformed to visualize the distribution of total acidity and firmness of red globe grapes. The results showed that the optimal prediction model for total acidity was MSC-CARS-SPA-PLSR, with correlation coefficient for the prediction set (Rp), root mean square errors of prediction (RMSEP) and residual predictive deviation (RPD) of 0.985 1, 1.348 2 and 5.664 3, respectively. The optimal prediction model for firmness was SG-CARS-PLSR, with Rp, RMSEP and RPD of 0.929 1, 7.935 4 and 2.510 8, respectively. In summary, hyperspectral imaging provides a new method for the detection and visualization of total acidity and firmness of growing red globe grapes.