基于贝叶斯神经网络的近红外光谱实木地板表面缺陷检测

实木地板的表面缺陷直接影响其力学性能和产品等级，表面缺陷的快速检测对实木地板的在线分选具有重要的现实意义。针对视觉方法检测实木地板表面缺陷识别率低的问题，提出了一种基于近红外光谱分析技术的检测方法。首先，分别采集规格为200 mm×100 mm×20 mm的表面带有活节、死节以及无缺陷的实木地板的光谱数据各60份，其中30份作为训练样本，30份作为测试样本；其次，使用高斯滤波(GSF)、分段多元散射校正(PMSC)和去趋势法(DT)等方法对采集到的光谱数据进行预处理，降低光谱噪声、消除光谱的散射影响；然后，利用改进遗传算法从处理后的光谱中提取特征波长用于构建缺陷识别与分类模型；最后，使用基于贝叶斯理论改进的神经网络构建实木地板缺陷识别和分类模型。实验使用含有活节、死节以及无缺陷的实木地板样本对模型进行训练和测试，结果表明：通过贝叶斯神经网络构建的缺陷识别与分类模型能够准确识别活节、死节和无缺陷三类实木地板，识别率分别为92.20%, 94.47%和95.57%。证明了实木地板表面缺陷类型与其近红外吸收光谱密切相关，并为下一步实现实木地板表面缺陷的准确定位提供快速检测方法。

Surface Defects Detection of Solid Wood Board Using Near-Infrared Spectroscopy Based on Bayesian Neural Network

Surface defects of solid wood boards directly affect their mechanical properties and product grade ,therefore ,to achieve rapid detection of surface defects has important practical significance for online sort of solid wood boards .In view of the low recognition rate of the surface defect of the solid wood boards ,a new method for the detection of 900~1900 nm was pro-posed .by using a portable near infrared spectrometer First of all ,the experiment collected absorption spectra of 180 samples with size of 200 mm × 10 mm × 10 mm ,consisting of 60 samples with live knots ,60 samples with dead knots and 60 defect-free sam-ples .Half of the samples were selected randomly as the training set ,and the rest of samples were test set .Secondly ,the the collecting NIR spectra of solid wood boards were preprocessed with Gaussian smoothing filter ,piecewise multiplicative scatter correction and De-trending to reduce the spectral noise and eliminate the influence of the scattering spectrum;Afterwards ,the improved genetic algorithm was utilized to select characteristic waves from the processed spectrum for building a model of defects recognition and classification;Finally ,a model for recognizing and classifying the defects of solid wood boards was built through the improved neural network based on Bayesian neural network .The experiments used three types ,containing live knots ,dead knots and defects free ,of solid wood board samples to train and test the model ,the results showed that the model of based on Bayesian neural network was able to accurately identify three kinds of them ,and the recognition rates were 92.20% ,94.47%and 95.57% ,respectively .This study demonstrates that the type of solid wood boards surface defects with its near-infrared ab-sorption spectra are closely related ,and the article provides a rapid approach to achieve the accurate positioning of solid wood board defects which is as the next step .