【作者】 吴富宁；

【导师】 朱虹；

【作者基本信息】 中国农业大学 ， 作物栽培与耕作， 2004， 硕士

【摘要】 本文基于计算机图象处理及相关信息技术的理论和方法，在盆栽和大田试验的基础上，结合常规观测手段，研究了采用田间近地数码相机图象诊断冬小麦氮素营养状况的可行性。建立了近地面数码相机图象获取和图象预处理方法，提取和筛选出了表征冬小麦不同氮营养状况下冠层颜色及形态的特征量，分析了小麦冠层颜色特征与叶片叶绿素含量和氮含量、叶片覆盖度与叶面积之间的相关性，提出了统计模式识别方法、叶色模型方法和应用叶片覆盖度三种氮营养诊断途径；通过实际检测，均获得了较高的正确诊断率。主要研究结果如下： 1、不同氮肥处理的冬小麦叶面积、干物重、上三叶叶绿素含量以及含氮量均存在一定差异，这一差异在低氮处理和中高氮处理间比较明显；孕穗追肥能够增加中低氮处理小麦后期的有效叶面积和干物重累积，提高上三叶的叶绿素含量和含氮量；拔节期和孕穗期上三叶叶绿素含量和氮含量变化呈正向线性相关，上三叶含氮量与植株含氮量呈高度线性相关；植株含氮量的变化范围小，能够用于表征小麦的氮素营养状况。 2、建立了数码相机图象的田间获取和预处理方法。本文提出的间隔距离提取法可以大幅度降低图象尺寸，并最大限度保持原始图象信息；采用比色订正基本消除了短时间间隔内（如一天）图象颜色信息随环境的变化，可以有效订正到10个灰度差异范围。 3、近地数码相机图象技术作为冬小麦氮营养诊断的方法，从图象中提取的颜色特征和叶片覆盖度获得了较好的诊断效果。 ①建立了应用模式空间分类方法进行冬小麦氮营养的诊断过程，一次施氮方式下，返青到蜡熟阶段，Have和Ⅰ综合光密度在低氮和中高氮处理间存在一个连续的差异曲线；采用两个特征进行平面空间分类，特征F₁（R-B，G-B）能够有效区分小麦拔节期和孕穗期低氮处理和中高氮处理情况，在小麦拔节期到开花期间可以明显区分低氮、中氮和高氮三类情况。 ②应用模式最小距离分类方法进行冬小麦氮营养诊断，建立了四个模式——构造[H]、综合光密度[I]、累加直方图[EG]、直方图距离[D]，对主要生育时期的30幅图象进行检测，在拔节、孕穗和开花期内中优9507和京411小麦均达到90％以上的正确分类率。 ③应用叶色模型进行冬小麦氮营养诊断，每个生育时期取60幅图象进行检测，起身到开花阶段正确分类率均在80％以上，灌浆和蜡熟阶段诊断效果最差；将Have和（R-B）作为各生育时期诊断模型的输入，Have特征的诊断效果略优于（R-B）特征，并在起身至开花阶段一直保持在80％以上的正确分类率，表明将Have特征作为叶色模型的输入进行冬小麦主要生育时期氮营养诊断是完全可行的。 ④应用叶片覆盖度进行冬小麦冠层图象的分类诊断，根据建立的诊断阈值，可以将起身期、拔节期、孕穗期、开花期和灌浆期的小麦图象分为严重缺氮、一般缺氮和中高氮三类；对两个品种各100幅图象进行检验，5个生育时期的正确分类率均在85％以上。 本文还对应用数码相机图象技术进行田间作物氮营养的诊断过程进行了方法和影响因素的探讨。更多还原

【Abstract】 According to the theory and method of the computer image processing thechnique and machine vision, it is proved that it is possible that N status of winter wheat is diagnosed by processing the digital machine images near field, based on field experiment and combined with general observation and test. The technique of capturing digital image and the method of image pretreatment are founded. The characteristics, which can express the canopy color of winter wheat under varied N-status, are abstracted and filtered. The relation between the color features of wheat canopy and chlorophyll content and nitrogen concentration of wheat body are analyized. In additon, the paper puts forward two means to diagnose N status: one is the color features diagnosis and the other is leaf covering features diagnosis. Two methods of image classification and identification, as well image colormodels, are included in the color features diagnosis. Through being practicly detected, the correct rateof diagnosis is high. Conclusions is drawed as following:1. There is obvious difference in the leaf area, dry matter and plant height of Getting up, jointing, booting and flowering phase between "Zhongyou9507" and "Jing411". The variety of chlorophyll content and nitrogen concentration of wheat canopy has also difference. The results show that the SPAD and nitrogen concentration of canopy have linear relation from turn green to dough stage, moreover, they consistently change with the difference of N supplied by soil.2. The effective images of wheat canopy are obtained by the digital camera, by the means of the limitative condition, combined with standard image pretreatment. The size of image can be decreased by the interval distance method, and the initial information of images can be preserved maximumly. The change of the colorful information of images with environment in shorter time can basically be eliminated by the means of color comparison, can be corrected to the range of ten gray value difference.3. The digital image processing technique could be used to detect N status of winter wheat. The better results of diagnosis can be obtained by abstracting color and leaf covering features from the images.(1) The method was established by Stat. classification features to diagnose N status of winter wheat. There is a continuous difference curve between low N treatment and middle-high N treatment to classify wheat canopy images by feature Have and MOD from the phase of returning to green to dough stage. Feature F1[(R-B),(G-B)] can effectively distinguish field wheat canopy images of those between low N treatment and middle-high N treatment during jointing phase and booting phase through combination of two features . And there is a clear division among low N , middle N and high N treatment using the combination feature from jointing phase to flowering stage.(2)Diagnose N status of winter wheat by the means of mode matching and build four matchingmodes--constructing [H], integrated optical density[1], added histogram[EG] and histogramdistance[D].Except trefoil stage, turn green and dough stage, the correct rate which wheat canopyimages of "Jing411" classify three parts: nitrogen absence, nitrogen middling and nitrogen abundance is over 73 percent, the correct classification rate of "Zhongyou9507" reaches 70 percent only in the Getting up, jointing, booting and flower stage, the correct classification rate of "Zhongyou9507" and "Jing411" in the jointing, booting and flowering stage is high and reaches over 90 percent.(3) Diagnose N status of winter wheat through leaf colour models. From the diagnosis results of sixty images of every growth season, the correct classification rate is over 80 percent in the flowering stage, but the result is not good in the grain filling and dough stage. The average values of H and (R-B) are used as the inputs of every growth season diagnosis models. The average feature of H is higher than the one of (R-B), and the correct classification rate is still over 80 percent from the getting更多还原