基于高光谱技术的灵芝孢子粉破壁率快速检测方法研究

目的 以破壁灵芝孢子粉为研究对象，利用高光谱技术结合化学计量学建立了灵芝孢子粉破壁率快速无损检测方法。方法 首先，采集不同破壁率灵芝孢子粉样品的高光谱图像，选定感兴趣区域后计算获得各样品可见-短波近红外波段(397-1004 nm)范围内的光谱数据；然后，比较了运用标准正态变量变换、多元散射矫正、Savitsky-Golay(SG)平滑、小波变换、SG平滑+标准正态变量变换及SG平滑+多元散射矫正等6种光谱预处理方法，竞争性自适应重加权、连续投影算法、无信息变量选择、最小角回归、遗传算法等5种特征波段提取方法以及偏最小二乘法、支持向量机、极限学习机、多层感知机及LightGBM等5种算法所建立的定量校正模型预测性能。结果 获得最优预测性能的算法组合为：SG平滑+竞争性自适应重加权特征波段选择+偏最小二乘；基于该算法组合建立的定量校正模型在破壁率区间为90%-100%的灵芝孢子粉样品预测集决定系数为0.8682、均方根误差为0.0117；进一步，将选定的最优算法组合应用于构建样品破壁率区间为0-100%的定量校正模型，计算测试集决定系数为0.9731，均方根误差为0.0493，表现出良好的泛化能力。结论 所建立的定量检测模型可以实现对灵芝孢子粉破壁率的快速、无损检测，为破壁灵芝孢子粉及其产品的质量控制提供了技术支撑。

A Rapid, Hyperspectral-based Method for Determining Sporoderm-broken Rate of Ganoderma Lucidum Spore Powder

OBJECTIVE To establish a rapid nondestructive detection method for the sporoderm-broken rate of Ganoderma lucidum spore powder by hyperspectral technology combined with chemometrics. METHODS Firstly, hyperspectral images of Ganoderma lucidum spore powder samples with different sporoderm-broken rates were collected, and spectral data in the visible-shortwave near-infrared band (397-1004 nm) range of each sample were calculated after selecting the region of interest. Then, we compared six spectral preprocessing methods (standard normal variable transformation, multivariate scattering correction, Savitsky-Golay (SG) smoothing, wavelet transform, SG smoothing + standard normal variable transformation, and SG smoothing + multivariate scattering correction), five characteristic band extraction methods (competitive adaptive reweighting, continuous projection algorithm, informationless variable selection, least angle regression, and genetic algorithm), and five algorithms (partial least squares method, support vector machine, extreme learning machine, multilayer perceptron, and LightGBM) for constructing quantitative correction models to predicts performance. RESULTS The optimal combination is SG smoothing + competitive adaptive reweighted feature band selection + partial least squares. The quantitative correction model established based on the algorithm combination achieved a prediction set coefficient of 0.8682, a root mean square error of 0.8682, and a root mean square error of 0.0117 for Ganoderma lucidum spore powder samples with a sporoderm-broken rate range of 90%-100%. The selected optimal algorithm combination is applied to construct a quantitative correction model with a sporoderm-broken rate range of 0-100%, the coefficient of determination for the test set is 0.9731 and the root mean square error is 0.0493, showing good generalization ability. CONCLUSION The established quantitative detection model can realize the rapid and non-destructive detection of the sporoderm-broken rate of Ganoderma lucidum spore powder, which provides technical support for the quality control of Ganoderma lucidum spore powder and its products.