桦木顺纹抗压强度的SEPA-VISSA-RVM近红外光谱预测

木材顺纹抗压强度是评价木材力学性能的重要指标,而传统测量方法操作复杂、精确度低。以桦木为例,提出基于近红外光谱技术(NIR)的SEPA-VISSA-RVM木材顺纹抗压强度模型,实现对其更加精确的预测。试验选取100个木材试件,在900~1700 nm近红外光谱波段上采集数据并测量抗压强度真值;然后采用卷积平滑(SG)方法进行光谱预处理;使用采样误差分布分析(SEPA)作为变量空间迭代收缩算法(VISSA)的改进策略进行特征波长优选;最后通过粒子群优化算法(PSO)优化核函数参数并建立相关向量机(RVM)的预测模型。试验表明:在特征波长优选方面,以偏最小二乘法(PLS)建模为基础的SEPA-VISSA方法,其预测决定系数为0.9593,预测均方根误差为2.8995,相对分析误差为3.0256,光谱变量数由512减小到111个,占总波长的22%,均优于VCPA、CARS和VISSA算法;在建模预测方面,以SEPA-VISSA所选波长为基础的RVM模型,PSO优化的拉普拉斯(Laplacian)核函数的核宽度为10.4043,决定系数为0.9449,预测均方根误差为2.0432,相对分析误差为4.2936,预测效果优于PLS和SVR。因此,基于近红外光谱的SEPA-VISSA-RVM建模能够实现对桦木顺纹抗压强度更准确和稳定的无损检测。

Prediction of compressive strength parallel to grain of birch wood using near infrared spectroscopy and SEPA-VISSA-RVM model

The compressive strength parallel to the grain direction of wood is an important index to evaluate the mechanical properties of wood,but the traditional measurement method is complicated in operation and low in accuracy due to the anisotropy and heterogeneity of wood.At the same time,the destructive detection method has the drawbacks of long time and high cost.Therefore,it is particularly important to accurately detect the compressive strength parallel to grain of wood in practical engineering applications.In this study,taking birch wood as an example,using the near infrared spectroscopy(NIR),the sampling error profile analysis-variable iterative space shrinkage approach-related vector machine(SEPA-VISSA-RVM)model was proposed for measuring the compressive strength parallel to grain of wood to achieve a more accurate prediction.In the experiment,the data collected at 900-1700 nm near infrared spectral band of 100 selected wood specimens were collected and were used to estimate the true value of compressive strengths.The Savitzky-Golay(SG)convolution smoothing method was used to preprocess the spectrum.SEPA was used as an improved strategy of VISSA to optimize the characteristic wavelength.Finally,the particle swarm optimization(PSO)was used to optimize the kernel function parameters and establish the prediction model of RVM.The experiment results showed that,in terms of characteristic wavelength optimization,the coefficient of prediction determination,root mean square error of prediction and relative percent deviation of the SEPA-VISSA method using the PLS modeling were 0.9593,2.8995 and 3.0256,respectively.The number of spectral variables decreased from 512 to 111,accounting for 22%of the total wavelength.All of them were better than those of VCPA,CARS and VISSA algorithms,and further improved the accuracy and robustness of the model for estimating the compressive strength parallel to grain of wood.In terms of modeling and prediction,for the RVM model based on the selected wavelength of SEPA-VISSA,the kernel width of the optimized Laplacian kernel function was 10.4043,the coefficient of prediction determination was 0.9449,the root mean square error of prediction was 2.0432,the relative percent deviation was 4.2936,and the prediction accuracy was better than those of PLS and SVR.Therefore,the SEPA-VISSA-RVM model using the near infrared spectrum not only had advantages in wavelength optimization,but also had better modeling accuracy than the commonly used PLS and SVR.At the same time,it could realize more accurate and stable nondestructive testing of birch compressive strength along grain,and had good applications in practical engineering aspects.