【作者】 刘镇波；

【导师】 刘一星； 王秀华；

【作者基本信息】 东北林业大学 ， 木材科学与技术， 2004， 硕士

【摘要】 应用先进的计算机彩色图像分析软件对木材横切面的细胞轮廓形态、细胞几何尺寸、组织比量等进行测定与分析；采用空间灰度共生矩阵方法提取木材横切面的显微图像灰度变化与构造特征参数，探讨了基于所提取的多种量化的图像特征参数实现树种判别的可行性，并初步建立基于木材横切面显微图像内容(特征参数)的木材树种判别程序。 对基于木材横切面显微图像所提取的33项特征进行主成分分析，结果表明前七个主成分因子集中了33个变量的90％左右的信息，它们分别是导管(树脂道)形态与尺寸因子、横切面纹理强弱及周期大小因子、组织比量因子、纹理粗细与复杂程度因子、木纤维(管胞)细胞的几何尺寸因子、木纤维(管胞)细胞形态因子、径向木纤维(管胞)细胞壁大小因子。同时结合变量间的相关分析，从33项特征中选取出了15项对树种判别具有实际意义的特征参数。 利用所选取出的15项特征参数，以最大相似原理初步建立了基于图像分析的计算机树种判别匹配算法，这种判别匹配算法是通过计算待判别树种与已知树种之间的相似系数来实现的。文中运用了最小差值差数判别法、树种综合特征阈值法、综合加权相似法三种方式进行相似系数的计算。其中最小差值差数判别法与树种综合特征阈值法是将每个特征参数对木材横切面显微图像特征的贡献率视为一致，只是树种综合特征阈值法是以阈值的大小来判断特征的相似性，而综合加权相似法是建立在对木材横切面显微图像特征参数的主成分分析的基础上，根据每个特征参数的贡献率大小计算相似系数。以相似系数计算为基础的计算机树种判别程序对未知树种的判别并不是给出唯一的答案，而是将数据库中的树种按相似系数的大小进行排序，同时能在主界面上显示树种的宏、微观特征和三切面显微图像，以便用户进一步验证待判别树种应为何种树种，提高了判别结果的准确性。 通过对木材横切面显微图像量化特征参数的提取与分析、基于图像分析的判别匹配算法和判别程序的建立，结果表明，以基于图像分析的方法达到木材判别的目的是基本可行的。此项研究结果将为木材鉴别、树木分类提供一种新的技术参考手段和初步的科学理论依据。更多还原

【Abstract】 It determined and analyzed the cellular contour, cellular dimension, tissue proportion, etc. of the wood transverse section by the software of multicolor image manipulation. And it determined the gray level and structure parameters of wood transverse section micrograph by spatial gray level cooccurrence matrix (GLCM). Then it researched on the feasibility of wood recognition according to the quantitative characters about micrograph, moreover, it programmed the wood recognition program based on the micrograph content of wood transverse section elementarily.It analyzed the 33 parameters about the micrograph of wood transverse section by the principal component and induced seven principal components. The results showed that the contribution of first seven principal components to total parameters was close to 90 percent. The first seven principal components were vessel(resin canal) morphological and dimension factor, texture intensity and period factor, tissue proportion factor, texture thickness and complexity factor, fiber (tracheid) dimension factor, fiber(tracheid) morphological factor, radial cell wall dimension of fiber(tracheid) factor, respectively. At the same time, it analyzed the correction among the parameters, and extracted 15 parameters which have significance for wood recognition from all parameters.It founded the computer aided wood recognition algorithm based on image manipulation according to 15 quantitative parameters. The algorithm was founded on the most similar principle. The key of the algorithm was computing the similar coefficient between the unknown sample and the samples in database. The methods of computing for similar coefficient were Discriminance of Minimal Difference of Parameters(DMDP), Discriminance of Limen of Tree Compositive Character(DLTC) and Synthetic Weight Similarity of Parameters(SWSP). The contribution of all parameters were considering as same for DMDP and DLTC, but it estimate s the similarity between parameters according to the limen for DLTC. DMDP was based on the principal component analysis for wood transverse section micrograph characters and computing the similar coefficient according to the contribution of every parameter. The recognition results of wood recognition program is not the only one. The all records in database are sorted by the similar coefficient. At the same time, the micro structural characteristic, the macro structural characteristic and the image of three sections are shown in the main interface. The user can testify which is the unknown sample. Then the veracity was improved.According to determined and analyzed of parameters of the wood transverse section micrograph, established of the recognition algorithm and programmed of wood recognition method, the results showed that the wood recognition method was feasibility based on the image manipulation. It would offer a new technical reference means and elementary theory foundation更多还原