为了实现木板材依据节子进行自动化分级,利用近红外光谱技术对针叶材表面节子进行检测。比较了光谱预处理和建模方法对节子识别的影响,研究了单一树种板材节子识别模型对其他树种板材节子识别的适应性,建立了混合树种板材的节子识别模型,并利用连续投影算法(SPA)进行了节子特征波长优选。结果显示,一阶导数光谱结合最小二乘支持向量机(LS-SVM)所建单一和混合节子识别模型性能最优。连续投影算法优选了15个特征波长变量,仅占全波长变量的0.87%,所建LS-SVM简化模型对测试集的敏感性、特异性和识别准确率分别为0.990,0.954,97.44%。实验结果表明,近红外光谱技术联合SPA与LS-SVM可以对多种针叶材板材的表面节子进行快速准确检测,连续投影算法是提取板材表面节子缺陷特征的有效方法,能简化模型并提高模型预测精度。 In order to realize automatic grading of wood based on knot, the surface knot of softwood was tested by near-infrared spectroscopy. The effects of spectral preprocessing and modeling methods on node identification were compared. The adaptability of single-tree plate node recognition model to the identification of other tree species nodules was studied. A nodal recognition model of mixed tree species plate was established. The projection algorithm (SPA) is optimized for knot characteristic wavelength. The results show that the first-order derivative spectra combined with least square support vector machine (LS-SVM) to build single and hybrid nodal identification model performance is optimal. The continuous projection algorithm optimized 15 characteristic wavelength variables, accounting for only 0.87% of the total wavelength. The sensitivity, specificity and recognition accuracy of the proposed LS-SVM simplified model were 0.990, 0.954 and 97.44% respectively. The experimental results show that near-infrared spectroscopy combined with SPA and LS-SVM can detect the surface knots of various softwood lumber plates quickly and accurately. The continuous projection algorithm is an effective method to extract the features of knot defects on the surface of plate, which can simplify the model Improve model prediction accuracy.