【作者】 马文杰；

【导师】 贺立源；

【作者基本信息】 华中农业大学 ， 植物营养学， 2007， 博士

【摘要】 烟叶的收购质量特征是评定烟叶等级的重要依据。目前，许多烟叶收购质量特征主要由人工靠经验获取，主观性强，且有许多指标呈模糊状态，往往出现评定级别不稳定的现象。在烟叶收购过程中还容易产生质量纠纷，不仅影响烟农的生产积极性，还给国家造成经济损失。鉴于目前人工分级存在的问题，用数字形式统一烟叶的收购质量特征是非常必要的。本文从烟叶数字图像采集技术规范、透射图像、颜色(形状)参照方法、烟叶光谱吸收规律等方面对烟叶收购质量的数字化特征进行了相关研究。摘要如下：1．首次系统研究了烟叶图像采集环境中照明电压、不同品牌数码相机、相机聚焦距离、相机分辨率和数码相机的不同拍摄模式、灯光随时间变化的稳定性等成像环境因素对图像识别结果的影响。首次在烟叶图像采集环境中引入颜色参照白板进行各种干扰因素相对扣除和物理尺寸绝对参照，比较了几种白板材料稳定性与干扰扣除效果，提出了相对颜色校正方法对因成像环境因素影响而引起的烟叶图像颜色识别误差进行校正。在此基础上规范了烟叶图像采集环境和操作步骤，对保障烟叶数字图像识别研究的质量控制和研究成果的共享具有深远的理论意义和实用价值。2．研究基于物质对光的吸收、反射和透射原理，在分析烟叶本身对透过不同波长光的吸收存在差异的基础上，专门设计并制作了能同时采集烤烟烟叶反射和透射图像的采集灯箱，发现并利用烟叶透射图像具有的特殊光谱吸收特征，提出了基于透射特性的烤烟图像背景分割技术。利用烟叶自身生物学特性对烟叶透射图像进行了区域统计法分割，结果表明分割精度高于传统点统计法，每片烟叶像素数平均高出0.65％，较好地保留了目标图像的原始特征信息。透射影像不仅可用于背景扣除，还有效扩充了烟叶数字图像处理的特征参数，为提升烟叶收购等级识别准确性奠定了基础。在数字图像处理中，利用生物自身特性进行背景扣除和扩充数字识别特征，具有专一性和排它性，对类似农产品数字识别研究具有广泛的指导意义。3．研究利用颜色参照白板的绝对参照作用实现烟叶的长度、宽度及面积等形状特征测量；用颜色参照白板标定烟叶图像的颜色色度变化，从中发现变化规律，为提取烟叶的分级特征奠定基础。通过人工方法测量烟叶长、宽度，称取烟叶单叶重；用计算机图像处理方法分别测出烟叶和颜色参照白板以像素为单位的长度和宽度，结合颜色参照白板的固定尺寸，通过变换计算得到烟叶实际的长宽度、面积，以及身份等分级指标特征；分别提取烟叶和同期采集的颜色参照白板在RGB和HSI两种颜色系统下的相关颜色特征，包括原始颜色和相对颜色。结果表明，利用参照方法测量的烟叶形状特征与人工测量结果之间有很好的相关性，11个供试级别烟叶的长度平均误差为2.59cm，相对误差平均值为5.11％；宽度平均误差为0.88cm，相对误差平均值为4.84％；身份特征测量结果也比较理想；相对色度可用来区分柠檬黄和桔黄色烟叶。因此，将图像处理技术与参照方法相结合，借助规则参照物标测不规则的烟叶形状、面积等特征参数是可行的，不仅测量方法简便，而且测量速度快；另外，用相对颜色方法区分烟叶的颜色特征也取得了一定的研究进展，为烟叶的数字化分级奠定了基础。4．研究烤烟烟叶在可见光区域的吸收规律，探索不同地区、不同级别烟叶在可见光区域的量化吸收特征，以及各吸收特征与烟叶地域性、级别属性之间的关系。利用改装并经过稳定性检测的GFY-160型荧光分光光度计，先使用320-810nm的波长范围，以5nm的波长间距对恩施、云南、临朐三个地区上、中、下不同部位的烟叶，以及同一地区上、中、下三个部位不同级别的烟叶进行透射，测得烟叶的透光强度，得到烟叶的透光特征曲线，根据透光强度的比值得到烟叶对光线的吸收特征。结果表明，烟叶在可见光范围内的确对蓝色光具有敏感的吸收作用；在获取的烟叶透射特征中，蓝橙比(％)(Blue／Orange，B／O)、绿橙比(％)(Green／Orange，G／O)随烟叶部位的变化规律比较明显，但蓝橙比(％)的变化规律强；红橙比(％)(Red／Orange，R／O)的变化规律不明显，不过上部叶的数值均在50％以上，表现出部分的规律性。因此，可以利用一些特殊波段下的吸收特征或者变换特征得到烟叶分级的量化特征。更多还原

【Abstract】 Tobacco purchase quality is an important reference for its grade. Currently, tobacco purchase quality is mainly estimated by human experiences that are sometimes subjective, and many factors are rather unclear, which leads to the unstable assessments of tobacco quality, results in the dissensions in tobacco purchasing, discouragement of farmers and the loss of state property. So it’s necessary to develop tobacco purchase quality regulations based on digital technology. In this dissertation, tobacco image collection regulations, transmission images, color (shape) references and the spectral absorptions of tobacco leaf were studied. The main resultes was summarized as follows:1. Environmental conditions such as lamp voltage, camera brand, focus distence, image resolution, shutting model and the stableness of lighting were studied to address the effects on the recognition of tobacco images. The color reference plate was introduced to eliminate the errors and calculate the absolute size of the tobacco. The stableness of the white reference plate was evaluated. The relative color was used to fast and timely correct the errors in tobacco image recognition. This study regulates the tobacco leaf collection environment. And laid foundations for the standardization of digital tobacco image collection framework.2. The tobacco leaves absorbance, reflection and transmission of visible light were studied. The special lamp box was designed based on the absorbance difference of transmitting light. The segmentation method of tobacco images was proposed based on the characteristics of the transmitting spectral. The results showed that the segmentation accuracy is higher than the method of point statistics. And the average pixels are 0.65% more than that of point statistics per leaf and the original information was well kept. The transmission image can not only trade off the background effects, and extend the characteristic parameters for tobacco images. The method presented maintains the objective characteristics of images, which help the increasing of segmentation accuracy and lay foundation for further studies on features of tobacco leaf. And also can serve as example for the recognition of other crop leaves.3. The measurements such as length, width, area and color of tobacco leaf were measured using the reference white plane. The mentioned measurements were measured by hand and by computer image processing method, respectively for the 11 tobacco leaf samples. The results showed there is a good relationship between the results obtained by different ways, and the average error for length is 2.59 cm, the corresponding relative average error is 5.11%, and the average error for width is 0.88 cm, the corresponding relative average error is 4.84%. The color characteristics of the tobacco leaf were also studied under RGB and HIS systems. It was concluded that the lemon and orange tobacco leaf can be distinguished by using the relative saturation. Based on the research mentioned above, it was found that in conjunction with the reference method, the image processing method could be used to extract the feature parameters of tobacco leaf such as shape and area, and relative color can be used to discriminate the color of tobacco leaf. The findings of the paper are very helpful for the digital rating for tobacco quality.4. The absorption characteristics of tobacco leaf in visible light were investigated to address the relationship between the tobacco origination, quality and classes. In this dissertation, the GFY-160 fluorescence spectrometer was employed to measure the intensity of the transmitted light at 5 nm intervals from 320 to 810 nm. The upper, middle and lower part of the tobacco leaf from Enshi, Yunnan, Linqu and from same area were conducted transmission experiment and the transmitted light were measured accordingly. The results revealed the tobacco leaf absorb blue light greatly, and in the transmission features, the ratio of blue and orange (B/O) and green and orange (G/O) change regularly with the different parts of the leaf while the ratio of red and orange (R/O) didn’t show the similar rule. And the R/O in the upper part of the leaf are all over 50%. The absorption features can be used to develop the quantitative rating systems for the tobacco leaf.更多还原