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

发布时间：2018-07-31 05:27

【摘要】：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200mm×100mm×20mm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。
[Abstract]:The surface defects of solid wood flooring directly affect its mechanical properties and product grades. The rapid detection of surface defects is of great practical significance to the on-line sorting of solid wood flooring. Aiming at the problem of low recognition rate of real wood flooring surface defects by visual method, a detection method based on near infrared spectroscopy (NIR) was proposed. First of all, the spectral data of 200mm 脳 100mm 脳 20mm with live knots, dead nodes and non-defective solid wood floors were collected respectively, of which 30 were used as training samples and 30 as test samples. In order to reduce the spectral noise and eliminate the spectral scattering effect, the Gao Si filtering (GSF), segmented multivariate scattering correction (PMSC) and the detrend method (DT) are used to preprocess the collected spectral data. The improved genetic algorithm is used to extract the characteristic wavelength from the processed spectrum to construct the defect recognition and classification model. Finally, the improved neural network based on Bayesian theory is used to construct the real wood floor defect recognition and classification model. The model was trained and tested with live, dead and non-defective wood floor samples. The results show that the defect recognition and classification model constructed by Bayesian neural network can accurately identify live nodes. The recognition rates of dead and non-defective wood flooring were 92.20% and 95.57%, respectively. It is proved that the type of surface defects of solid wood flooring is closely related to its near infrared absorption spectrum, and a fast detection method is provided for realizing the accurate location of solid wood floor surface defects in the next step.
【作者单位】：东北林业大学机电工程学院;东北林业大学材料科学与工程学院;
【基金】：国家林业局948项目(2015-4-52) 中央高校基本科研业务费专项资金项目(2572017DB05)资助
【分类号】：O657.33;TS67

【相似文献】