温度变化对近红外光谱预测木材含水率的影响

为实现温度不稳定环境下木材含水率的近红外光谱检测，探究了不同温度下木材近红外光谱的变化规律及温度变化对近红外预测木材含水率的影响。对从林场采集的樟子松、水曲柳、大青杨和红松原木木块试样各75块，共计300块试样，进行了不同温度和含水率条件下的近红外光谱采集。采用单一温度下的校正集分别与各个温度下的验证集建立偏最小二乘含水率预测模型，探究温度变化对木材含水率模型预测准确性的影响。比较了不同光谱预处理的木材含水率预测温度全局模型。采集相同含水率下不同温度的近红外光谱数据，对光谱进行光谱平均、一次微分、主成分分析和偏最小二乘判别分析，以探究温度变化时，木材近红外光谱的变化规律。结果显示：（1）温度对木材样品光谱存在显著影响；主成分分析和判别分析表明不同温度下的样品有明显聚类趋势，温度判别准确率为96.1%。温度会影响木材的近红外光谱在特定波长吸收峰的位置及吸光度，在含水率相同的情况下，随着温度的升高，特定位置吸收峰有逐渐向高频波段转移的趋势且在零下低温时波峰移动变化更明显。（2）不同温度下的PLS含水率预测模型对温度变动的适应能力有差异，木材含水率预测模型更适应于检测与建模样本相同温度的样品。与单一温度模型相比，PLS温度全局模型对于温度变化具有很好的适应性和应用潜力，RMSEP低于大部分单一温度模型。基于SG平滑+多元散射校正+一次微分预处理联用的PLS含水率温度全局模型有较好的预测效果和温度适应性，RMSEP降为0.074。可见，温度变动是近红外法检测木材含水率的过程中不可忽视的扰动因素；基于光谱预处理的温度全局模型可以显著提高温度适用性。该研究可进一步促进近红外光谱技术在木材生产、加工过程中的应用。

Influence of Temperature Change on the Prediction of Wood Moisture Content by NIR

To accomplish the near-infrared spectroscopy detection of wood moisture content under an unstable temperature environment, the change law of wood near-infrared spectroscopy under different temperatures and the influence of temperature changes on the near-infrared prediction of wood moisture content were explored. Using 75 pieces of log samples of Pinus sylvestris, Fraxinus mandshurica, Populus sylvestris, and Korean pine logs collected from the forest farm, a total of 300 pieces of samples were used to conduct near-infrared spectroscopy under different temperature and moisture content conditions. The correction set at a single temperature was used to establish a partial least squares moisture content prediction model with the verification set at each temperature. The influence of temperature changes on the prediction accuracy of the wood moisture content model was explored. The global models of wood moisture content prediction temperature based on different spectral pretreatments are compared. Collect infrared spectroscopy data at different temperatures under the same moisture content, and perform spectral averaging, differential observation, principal component analysis, and partial least square discriminant analysis on the spectra to explore the law of wood near-infrared spectroscopy changes with temperature. The results show: (1) Temperature significantly affects the spectrum of wood samples. Principal component analysis and discriminant analysis show that samples at different temperatures have a clear clustering trend, and the accuracy of temperature discrimination is 96.1%. The temperature will affect the position and absorbance of the absorption peak at a specific wavelength in the near-infrared spectrum of wood. With the same moisture content, as the temperature increases, the absorption peak at a specific location tends to-shift to the high-frequency band gradually and is at a sub-zero temperature when the peak movement changes more obviously. (2) The PLS moisture content prediction model at different temperatures has different adaptability to temperature changes. The wood moisture content prediction model is more suitable for detecting samples at the same temperature as the modeling sample. Compared with the single temperature model, the PLS temperature global model has good adaptability and application potential for temperature changes, and the RMSEP is best at 0.082. The PLS water content temperature global model based on SG smoothing + multivariate scattering correction + first-order derivative preprocessing has a better prediction effect and temperature adaptability, and the RMSEP is reduced to 0.088. It can be seen that temperature variation is a disturbance factor that cannot be ignored in detecting wood moisture content by the near-infrared method. The global temperature model based on spectral pretreatment can significantly improve the temperature applicability. This research can further promote the application of near-infrared spectroscopy technology in wood production and processing.