【作者】 张瑜；

【导师】 宋文龙； 韩玉杰；

【作者基本信息】 东北林业大学 ， 机械设计及理论， 2012， 博士

【摘要】 根系是植物从土壤中获取营养的重要器官,其生长状况对植物有着至关重要的影响,而且根系通过与土壤形成复合结构体的方式,起到了固定植物地上部分以及固定土壤防止水土流失的作用,同时在生态系统循环中根系的碳汇作用也是不可忽视的。由此可知对根系进行研究的意义是非常重大的,但是由于根系隐藏于地面以下,很难对根系进行直接的观察。微根窗技术的提出为人们对植物根系的研究带来了极大的方便。本论文在国家自然科学基金资助项目(30972424/C0414)的支持下,对微根窗技术获得根系图像的处理技术进行研究,提高了根系图像处理的速度与精度,同时还在此研究基础上对植物的根系形态参数测量方法进行了分析与研究。本文的主要研究内容有以下几个方面：论文提出使用模糊算法对噪声进行分类,将噪声分为高斯噪声、处于边缘的脉冲噪声以及处于图像平坦区域的脉冲噪声,分别采用模糊加权均值滤波、双向多级中值滤波和单向多级中值滤波的方法进行滤波处理,自适应分类滤波算法去除图像噪声的同时较好地保护了边缘细节。对微根窗采集到的根系图像进行图像增强及去噪,减小原始模糊图像边缘的宽度,为后续图像处理做准备。通过图像拼接将微根窗获取的多幅根系局部图像拼接为完整的根系图像,以获得较为全面的根系形态分布。本研究提出相位相关法与特征点匹配相结合的方式进行图像拼接。改进后的Harris角点检测算法提高了对灰度变化的敏感性及定位的准确性；改进角点响应函数解决了原有函数中K值设定的随机性；根据首图像处理结果自动设定后续图像角点响应函数的阀值T；对完成匹配的图像进行亮度调节。论文中根据Canny三准则选择三次B样条小波函数进行自适应阀值多尺度根系图像边缘特征提取,并将检测后的多幅图像进行数据融合得到准确的根系边缘特征图像。通过对数学形态学的开闭操作进行根系的形态分布及参数测量。利用膨胀和腐蚀等技术对所提取的根系边缘特征图像的毛刺、凹陷、间断及孤立的小孔进行处理,利用数学形态学的薄化运算对根系边缘图像进行细化,为后续根系形态参数测量提供数据来源。根据图像像素与实际尺寸存在的线性关系及根系形态参数的几何性质进行根系的长度、表面积、平均直径、体积以及根系间的夹角等形态参数测量。本论文通过对微根窗获取的根系图像进行增强及去噪,图像拼接,根系边缘特征提取及形态参数测量,实现了根系图像的精确采集及测量,为植物根系重构以及后续的固土机理研究,碳汇作用研究及气象预报方面提供了详实准确的数据来源。更多还原

【Abstract】 Root is vital organ for plants to get nutrition from the soil, if it grow well has crucial effect on the whole plant, and root can form a composite structure with the soil to fix plants and to prevent soil erosion. At the same time, carbon sink of root in the ecosystem circulation can not be ignored. So the study on root is of great significance. But root is hidden, and is difficult to directly observe. Invention of minirhizotron gives people great convenience to do research on root. This article focused on image processing of root that achieved from minirhizotron, and improved the speed and accuracy of processing. After that, we also did some analysis and research on root morphology parameter measuring methods. In this paper, the main contents are as below:In this paper, we enhance and denoise root image collected by minirhizotron, reduce edge width of original fuzzy image, and prepare for the subsequent image processing. This study presents a fuzzy algorithm to classify the noise. Noise is divided into the Gauss noise, pulse noise that at the image edge and impulse noise in flat image regions, use the fuzzy weighted average filter, bidirectional multilevel median filter and the one-way multilevel median filtering method separately for filtering. Adaptive classification algorithm can not only remove the image noise, but also give the edge details better protection.We can obtain a complete root image through mosaicing local minirhizotron image, and then obtain a more comprehensive morphological distribution. This study presents the phase correlation and feature point matching for image mosaic. The improved Harris corner detection algorithm improves the Gray’s sensitivity and positioning accuracy of image; the improved corner response function helps to solve the random setting of K value of original function. According to the processing results of forward image, system can set threshold T of the follow-up image corner response function automatically, and can regulate brightness of the completed matching image.This article also did research on root edge feature extraction algorithm of the joined image. We chose cubic B-spline wavelet function according to Canny three guidelines for adaptive threshold multi scale image edge feature extraction of root, and obtain accurate root edge characteristic image based on multiple image data fusion after detected.We get root morphological distribution and parameter measurement through open and close operation of mathematical morphology. The study uses dilation, erosion and other techniques to extract the root edge feature like burring, depression, discontinuous and isolated holes, thins the root edge through mathematical morphology thinning operations, and provides data sources for the following root morphology parameters measurements. According to the linear relationship between the image pixels and actual size, and geometric properties of root morphological parameters, we can get the morphological parameters measurement like root length, surface area, mean diameter, root volume and angle.Through enhancement, denoising, image mosaic, root edge feature extraction and morphological measurement parameters of minirhizotron root image, we can obtain accurate acquisition and measurement of the root image, and provide detailed and accurate data foundation for research on plant root remodeling, subsequent soil reinforcement mechanism, carbon sequestration effect, and forecast.更多还原