应用近红外光谱分析对竹材力学强度的预测

Mechanical Strength Prediction of Bamboo Using Near Infrared Spectroscopy

使用900~1700 nm近红外光谱分析仪对展平竹进行多点采样,随后利用深度置信网络(DBN)与局部支持向量回归(LSVR)进行训练,建立DBN_LSVR预测模型来预测单元材料的力学性能。采集到的光谱数据经过S-G滤波,完成了数据平滑,去除了大部分噪声。DBN-LSVR预测模型的竹材断裂模量(MOR)和弹性模量(MOE)的均方根误差分别为716.23、17.54,平均绝对误差分别为524.40、16.82。将预测误差与BP、DBN、LSVR模型预测误差比较,DBN_LSVR模型的预测误差最小。基于DBN_LSVR的MOR和MOE预测模型的相关系数分别为0.85和0.80。提出的方法可以有效地对展平竹力学强度进行无损分析,缩短了判别竹材物理性能的时间和成本。

With near infrared spectroscopy(NIRS),a nondestructive and fast method for mechanical strength prediction of bamboo was studied.A 900-1700 nm NIR analyzer was used to conduct multi-point sampling on flattened bamboo.Then,deep confidence network(DBN)and local support vector regression(LSVR)were used for training,and a DBN_LSVR prediction model was established to predict the mechanical properties of the unit material.The collected spectral data were smoothed by S-G filtering,and most of the noise was removed.The root mean square errors of bamboo’s MOR and MOE based on the DBN_LSVR prediction model are 716.23 and 17.54,respectively,the mean absolute errors are 524.40,16.82,respectively.Compared with BP,DBN,LSVR model prediction error,DBN_LSVR model prediction error is the smallest.The correlation coefficients of MOR and MOE prediction models based on DBN_LSVR are 0.85 and 0.80,respectively.The proposed method can effectively analyze the mechanical strength of flattened bamboo without damage,and shorten the time and cost of distinguishing the physical properties of bamboo.