【作者】 王阿川；

【导师】 曹军；

【作者基本信息】 东北林业大学 ， 机械设计及理论， 2011， 博士

【摘要】 单板质量好坏直接影响了用单板制成的人造板材的强度及表面质量和等级,为了提高单板的等级和木材的利用率,目前通常采用人工对单板进行缺陷检测及缺陷挖切修补,自动化水平低,劳动强度大、生产效率低,严重影响了经济效益,增加了生产成本。将机器视觉和机器人技术引入生产,将可以有效地克服人工检测修补所带来的缺点,对于提高我国人造板行业自动化水平起到很好的推进作用,具有重要的学术意义和应用价值。本文基于变分和偏微分方程(PDE)的图像处理、图像修复技术及机器人技术,结合单板的特点,对单板表面的缺陷进行有效识别和修复,形成一种单板节子的快速检测、缺陷挖切和修补方法。论文主要研究内容与工作如下：本文主要对C-V模型进行了改进,解决了C-V模型在多目标分割以及复杂背景情况表示上的局限,以适应单板缺陷图像的多目标分割。首先,将背景填充技术与改进C-V模型及AOS半隐式方法相结合,提出了基于AOS格式的多相改进C-V模型及背景填充耦合的单板缺陷分割算法,解决了单板缺陷图像、木材缺陷图像的多目标自动分割问题。第二,由于现有的图像采集系统所获取的多为矢量或彩色图像,针对单板矢量或彩色图像缺陷分割问题,提出了基于AOS的多相改进矢量C-V模型及背景填充的单板缺陷矢量图像分割方法,将单板矢量图像作为一个整体图像进行处理,实现了单板缺陷彩色图像的多目标分割问题。第三,针对带纹理单板缺陷彩色图像,结合多通道Gabor滤波器、改进矢量C-V模型,提出了多相改进矢量C-V模型与Gabor滤波器的单板缺陷彩色图像分割方法,解决了带纹理单板缺陷矢量图像的多目标识别问题,得到了识别结果与原图像相同的分割图像,并可生成单板缺陷修补的彩色掩膜图像。针对各种形状的单板节子缺陷,特别是带有凹形区域的节子,以及单板节子缺陷目标和背景颜色相近、边缘不清晰等造成缺陷识别困难的多目标识别问题,本文结合了基于边缘的活动轮廓模型和基于区域的活动轮廓模型,提出了一种基于全局最优的活动轮廓模型的多目标检测方法,通过利用对偶形式的数值计算方法,减小了计算量,提高了分割速度,实现了对复杂纹理背景下的单板多节子目标的有效检测。针对含有较丰富纹理的单板缺陷图像,本文采用了变分偏微分方程的图像分解方法进行单板缺陷检测,首先,在ROF模型的基础上,结合高阶导数的图像分解模型,提出了一种消除阶梯效应的单板缺陷图像分解模型,运用半二次规整化方法求解该模型,得到了分解单板缺陷图像的有效方法,保护了结构图像的边缘,更好的提取纹理特征。其次,结合AAFC模型与TV正则项一般式,提出了一种联合图像结构-纹理分解和边缘检测耦合的图像分解模型。实现了在进行单板缺陷图像结构-纹理部分分解的同时,又得到了较好的单板缺陷边缘检测结果。为了将图像修复理论、方法应用于单板表面节子缺陷图像的自动修补中。提出了一种BSCB改进算法,使其在非纹理单板节子图像修复上得到了比较好的效果。针对单板节子区域较大、节子周围纹理比较复杂的情况,又提出了将BSCB改进算法与基于样本块的图像修复算法相耦合的单板缺陷图像修复新方法,实现了对单板节子缺陷图像的修复,达到较好的修复效果。针对单一图像修复方法的局限性,提出了一种基于图像分解的单板缺陷图像修复方法。首先,改进了VO模型实现了对单板缺陷彩色图像的有效分解,得到了单板缺陷图像的结构部分与纹理部分；然后,采用BSCB改进算法,对单板缺陷图像结构部分进行修复；采用基于样本的Criminisi算法对纹理部分进行修复；最后,将修复好的图像叠加合成,达到比较好的修复效果。最后,提出了基于机器视觉的机器人单板缺陷检测修补系统的设计方案,并对系统组成、工作原理加以分析。更多还原

【Abstract】 Veneer quality has a directly influences on the strength, the surface quality and level of man-made sheets which are made by veneer, at present, in order to improve the level of veneer and the rate of wood utilization, we usually use artificial methods to detect veneer defects and cut or repair veneer defects, because of low level of automation, labor-intensive and low productivity, which have a serious impact on economic benefits and also increase the production cost. So bringing the machine vision and robotic technology into production can be effectively overcome the shortcomings which caused by artificial repair, and can improve the level of automation of the wood-based panels industry, which has an important academic significance and application value. The paper is based on the image processing of variational and partial differential equations(PDE), image restoration techniques and robotics, combining the characteristics of veneers they can have an effective identification and repair about the defects of veneer surface and then form the methods of rapid detection,digging and patching of knots. The main contents are as follows:In this paper, it improves the model of the C-V and solves the limitations about multi-objective segmentation and complex background to fit the multiple objects segmentation of veneer defect image. First, combining the background filling technology, improved C-V model with the AOS semi-implicit method, the paper puts forward segmentation algorithm of multi-phase level veneer defect which based on improved C-V model of AOS and the coupling of background filling,solves the multi-objective automatic segmentation problem of veneer or wood images which have defects. Second, as the collection images from systems are mostly vector images or color images, about veneer vector or color image for defect segmentation problems, it puts forward image segmentation method of veneer defect vector which based on improved multiphase vector C-V model of AOS and background filling, process veneer vector image as a whole and achieve multi-objective segmentation of veneer defect color image.Third, about the defect color image of the veneer with texture, combining multi-channel Gabor filter with improved vector C-V model,it puts forward color image segmentation method of veneer defect which based on improved vector C-V model and Gabor filtering,it also solves the multi-objective identification problem and get segmentation of image which recognition results and original image are the same, therefore generates repaired color mask image of veneer defect.About the defect of veneer knot for a variety of shape, especially the knots which have concave areas and the background color is similar to the color of veneer defect,the edge is unclear such kind of multi-objective recognition problems which caused recognition difficulties, in this paper, it combines the active contour model which based on edge and region, and puts forward multi-objective detection method of active contour model based on a fast global minimization and then the use of the numerical calculation in the dual form, it greatly reduces the calculation and improves the segmentation speed and realizes effective testing of the veneer knot objects under the complex texture background.For the defect veneer image with rich texture, the paper uses the image decomposition method based on variational and PDE to inspecting the defect of veneer. First, based on the ROF model, combining image decomposition model of higher derivative, the paper puts forward a veneer defect image decomposition model of eliminating ladder effect, using the Half-Quadratic Regularization method, we solve this model and get effective method of decomposition of veneer defect image, therefore protect the edge of the structure image and can better extract the texture features.Second, combine the AAFC model and TV with the general form, the paper puts forward a new model of decomposition which joints image structure and texture decomposition and coupling of edge detection. The paper decomposes the part of image structure and the texture of defect veneer image effectively, at the same time get a better edge detection result.In order to apply image restoration theory and methods to the automation repair of the veneer knots on the surface defect images. The paper puts forward an improved BSCB algorithm, and gets a relatively good result of repairing veneer knot images of non-textured.For veneer knots which is larger regional or texture is more complex,the paper puts forward a new restoration algorithm of veneer defect image which based on coupling improved BSCB algorithm and image restoration algorithm of sample block, and realizes restoration for veneer defect image,and achieves a better restoration result.For the limitations of single image restoration method, the paper proposed a image restoration method of veneer defects based on image decomposition. First, improved VO model and realized effectively decomposing the color image of veneer defects, got the texture and structure parts of defect veneer images. Then, adopted the improved BSCB algorithm and repaired the structure parts of veneer images; the paper selects Criminisi algorithm based on the sample for repairing the texture parts. At last, overlay and synthesis the restored images to achieve better restoration results.At last, the paper puts forward a design scheme and develops the robot veneer defect detection and repair system which is based on the machine vision, then it analysis system component and working theory.更多还原