厚度为0.8 mm的超薄纤维板是目前纤维板品类中的试验创新产品, 树皮含量对其生产设备参数的设定以及静曲强度、 耐水性等质量指标影响较大, 精确测定超薄纤维板木纤维中树皮含量极为重要。 目前树皮含量的精确测定较为困难, 本试验通过高光谱近红外成像系统结合相关算法建立了纤维树皮含量检测模型, 创新了纤维树皮含量的检测方法。 利用高光谱成像仪分别测定了含有杨木树皮为0%、 3%、 5%、 7%、 10%、 12%、 15%、 20%、 25%、 30%和100%的杨木纤维样本的光谱图像。 分析了采用均值中心化(MC)、 多元散射校正(MSC)、 标准正态变量变换(SNV)以及一阶(1-Der)导数四种预处理的对比结果, 从而选择最优预处理方法为MSC。 对MSC预处理后的光谱数据采用SPA及CARS进行特征波长提取, 得到与树皮含量相关性最高的波段组合, 并与全波段模型进行对比分析, 建立偏最小二乘回归(PLSR)模型。 从实验数据可以看出: MC, MSC, SNV 和1-Der四种预处理建立的偏最小二乘回归(PLSR)模型预测性能存在差异, 其中全波段MSC-PLSR模型的性能最好, 其校正决定系数$R_{\mathrm{C}}^{2}$为0.994, 预测决定系数$R_{\mathrm{P}}^{2}$为0.985, 校正均方根误差RMSEC为0.831%, 预测均方根误差RMSEP为1.336%。 通过SPA和CARS分别提取了37个和49个特征波段, 其中CARS模型更好, 其$R_{\mathrm{C}}^{2}$值为0.991, $R_{\mathrm{P}}^{2}$值为0.979, RMSEC值为0.885%, RMSEP值为1.335%。 实验结果表明: 高光谱成像系统结合相应算法可以实现对纤维树皮含量的检测, 该研究结果为超薄纤维板生产中树皮含量的检测提供了技术支持和理论参考, 可以有效实现纤维中树皮含量的定量检测, 创新建立了一种能够测定纤维板树皮含量的模型方法。

Ultra-thin fiberboard with the thickness of 0.8mm is an innovative experimental product in the fiberboard category. The bark content greatly influences the setting of its production equipment parameters and the quality indicators such as static curvature strength and water resistance. It is important to determine the bark content in ultra-thin fiberboard wood fiber accurately. At present, the accurate determination of bark content is difficult. A fiber bark content detection model was established by hyperspectral near-infrared imaging system combined with relevant algorithms, and the fiber bark content detection method was innovated. In this experiment, spectroscopic sample images of poplar fibers containing poplar bark of 0%, 3%, 5%, 7%, 10%, 12%, 15%, 20%, 25%, 30%, and 100% were determined by the hyperspectral imager. The results of pretreatment of mean centralization (MC), multiple scattering corrections (MSC), standard normal variable transformation (SNV) and first-order (1-Der) derivative were analyzed, then the MSC was selected as the best pretreatment method for this test model. The spectral data pretreated by MSC were extracted by SPA and CARS, and the band combination with the highest correlation with the bark content was obtained, and the full-band model was compared and analyzed to establish a partial least squares regression (PLSR) model. From the experimental data, we can see differences in the predictive performance of the model of partial minimum secondary return (PLSR) established by pretreatment of MC, MSC, SNV and 1-Der. Among them, the performance of the MSC-PLSR model is the best. The correction determinant $R_{\mathrm{C}}^{2}$ is 0.994, the prediction determinant $R_{\mathrm{P}}^{2}$ is 0.985, the correction square root error RMSEC is 0.831%, and the prediction square root error RMSEP is 1.336%. 37 and 49 characteristic bands were extracted by SPA and CARS respectively, among which the CARS model was better, $R_{\mathrm{C}}^{2}$ was 0.991, $R_{\mathrm{P}}^{2}$ was 0.979, RMSEC was 0.885%, and RMSEP was 1.335%. The experimental results show that the hyperspectral imaging system combined with the corresponding algorithm can realize the detection of the bark content of the fiber. The study's results provide technical support and theoretical reference for the detection of the bark content in the production of ultra-thin fiberboard, which can effectively realize the quantitative detection of the bark content in the fiber, and innovate a model method that can determine the bark content of the fiberboard.