基于视觉显著性的木板实时分类方法研究  

Real-time classification of wood based on visual significance

作　　者：李晖[1] 吴佳宁 苑玮琦[1] 隋春江 Li Hui;Wu Jianing;Yuan Weiqi;Sui Chunjiang(School of Information Science and Engineering,Shenyang University of Technology,Shenyang 110870,China)

机构地区：[1]沈阳工业大学信息科学与工程学院,沈阳110870

出　　处：《仪器仪表学报》2018年第12期237-244,共8页Chinese Journal of Scientific Instrument

摘　　要：木板分类是木制家具制作的重要环节,现有的木板分类算法仅从纹理特征方面进行分类,且运用的纹理分析方法与实际人眼分类标准具有一定的差异性,如局部二值模式法(LBP)、灰度共生矩阵法(GLCM)等方法。从人眼仿生角度出发,将视觉显著性融入基于木板表面综合特征的分类算法中,提出一种基于木板视觉显著性的分类方法。采用高精度线阵相机搭建机器视觉系统进行木板图像的实时采集,通过动态阈值分割、特征筛选、形态学处理等方法识别图像中的木板区域,提取基于视觉显著性的木板纹理区域占空比,木板纹理区域与背景区域的对比度等特征,构建多层神经网络进行分类识别。利用从木材加工厂实时采集的1 156张木板图片进行分类实验,实验证明基于视觉显著性输入特征的多层神经网络可以完成木板分类任务,且具有94.17%的分类准确率。Classification of wood is an important part in wooden furniture production. The current wood classification algorithms only classify the texture features, and texture analysis methods have certain differences from the actual human eye classification standards, such as LBP, GLCM methods. From the perspective of human eye bionics, this paper proposes a classification method based on the visual saliency of the wood, to integrate visual saliency into the classification algorithm based on the comprehensive characteristics of the wood surface. A machine vision system is built with a high-precision linear array camera to collect real-time image of wood. The wood region in the image is identified by dynamic threshold segmentation, feature screening, and morphology processing, Extracting features are used to model multilayer neural network for classification and identification, which include the duty ratio of the wood texture area, the contrast between the wood board texture area and the background area based on visual saliency. Using 1 156 wood images collected from wood processing plants in real time, the experiments are conducted to demonstrate that multi-layer neural networks based on visually significant input features can accomplish wood classification tasks with a classification accuracy of 94.17%.

关 键 词：木板 机器视觉 纹理分割 视觉显著性 多层神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TH865[自动化与计算机技术—计算机科学与技术]