胚芽鞘是作物幼嫩子叶尽快摆脱初期恶劣环境的保护组织.细胞壁的松弛是胚芽鞘伸长的前提,而β-葡聚糖酶及膨胀素等是打开半纤维素连键的关键松弛酶(蛋白).芽鞘伸长生长过程中,在生理生态水平上有较大变化,IAA起主要的调节作用,同时光、温、水等生态环境因子的影响,引起芽鞘内部生理上发生一系列相应的变化.本文综述了胚芽鞘伸长的分子机理及生理生态应答,并探讨了在实践中的应用价值.

It is an important method to improve the planting profit of Phalaenopsis ababilis by the control of its anthesis through image identification technology to acquire the growing parameters of greenhouse Phalaenopsis ababilis. How to segment and extract the image of Phalaenopsis amabilis from the background image is the key point of image identification. This paper firstly uses color gradient algorithm to extract the Phalaenopsis amabilis natural images gradient image. It adopts the threshold value to find out the significant part of gradient image (e.g. the significant edge in the image), and the watershed segmentation method to segment the source image. It uses the significant edge image to determine the watershed segmentation image so as to remove the insignificant "dam" and make its both sides converge, to restrain over-segmentation effect. Finally, the target image of Phaleaenopsis amabilis will be obtained through merging the similar areas according to area merging standards. By comparing the segmentation effect with artificial method of the 20 Phalaenopsis ababilis images, it proves that the segmentation algorithm based on gradient and watershed method has good effect to extract the Phalaenopsis amabilis image from its natural background and the segmentation rate is up to 93.6%.

利用图像识别技术采集设施蝴蝶兰生长参数，从而对花期进行调控，是提高蝴蝶兰种植效益的重要手段，而如何把蝴蝶兰图像与自然背景图像进行分割与提取，是图像识别的关键。该文利用彩色梯度算法，提取出蝴蝶兰自然图像的梯度图像，利用阈值找出梯度图像的显著部分（即图像中的显著边缘），同时利用水域分割方法对源图像进行分割，产生过分割图像，然后利用显著边缘图像对过分割图像进行判断，去除不显著的“水坝”并令其两边水域相融合，从而达到极大的抑制过分割的目的，最后再根据区域合并准则合并相似的区域，得到蝴蝶兰目标物图像。针对20幅蝴蝶兰图像，通过与人工分割的方法进行对比试验，结果表明，该文提出的基于梯度和分水岭的分割算法能够很好地从自然背景中提取出蝴蝶兰图像，分割正确率达到了92.6%。

摘　要： 提出了利用归一化的红绿色差（R-G）/（R＋G）分割苹果的方法。对不同光照情况下拍摄的苹果图像进行了识别,并对识别后的图像进行预处理后,获得苹果的轮廓图像。对轮廓图像采用随机圆环法进行果实圆心、半径提取。通过建立基于面积特征与极线几何相结合的匹配策略实现双目视觉下的果实定位,对于搜索区域内面积相似的果实,通过计算垂直投影的互相关函数最大值的方法,得到排序基准线,然后根据顺序一致性原则进行匹配。实验结果表明：识别算法可以较好地消除阴影、裸露土壤等影响,识别率达到92%。采用随机圆环法,可以准确地提取果实的圆心、半径。在60~150 cm的距离范围内,测量误差小于2 cm。

采用基于亮度的阈值分割算法对G-B灰度图像进行分割,采用4连通像素标记法对果实目标进行标记,提取质心、面积、外接矩形、切断点等特征。采用平面标定方法,建立摄像机矫正数学模型,利用Matlab多次标定计算完成双目视觉系统进行自标定。根据图像分割结果,选取质心为匹配基元,采用极线约束、视差梯度约束的算法来实现果实目标的匹配。根据相似三角形原理,计算茄子果实深度信息。结果表明,测量距离300~600 mm范围内,茄子深度信息的误差基本上在±15mm以内,平均用时0.26 s。基于双目视觉的茄子采摘机器人识别与定位方法原理简单,智能性好,适应面广,能够满足采摘机器人目标识别和定位的要求。

For the intelligent tea picking machine, the positioning of tea tips was a difficult procedure. A positioning method based on active computer vision and corresponding visual system were proposed to solve this problem. First, according to the characteristics of the tea picking surface, a cross light path of projection and camera was designed; then the recognition approach of tea tips based on the color factor was developed in natural environment; furthermore, fringe projection profilometry was studied to acquire the height information of tea tips. In fringe projection profilometry, the temporal phase unwrapping method was used to obtain the phase maps, the invalidity identification framework and morphological filter were designed to remove noise points, and a polynomial approximation method which can reduce the nonlinear error was applied to calculate the height. The experimental results show that the positioning system for tea tips based on fringe projection technology can effectively recognize the tips and extract their three dimensional information, which can help new tea picking machine to realize the positioning function.

为解决名优茶智能采摘的茶叶嫩梢定位问题，提出了一种基于主动视觉的嫩梢定位方法，并据此开发了一套视觉定位系统。针对茶叶采摘工作面特点设计了交叉投影与摄像光路；基于色彩因子法实现了自然环境下茶叶嫩梢的识别；重点研究了利用光栅投影轮廓术实现茶叶嫩梢高度信息获取。光栅投影轮廓术采用时间相位法获取光栅场的相位，并利用判别机制和形态滤波器去除噪声点，以多项式逼近法计算高度克服非线性误差。试验结果表明，基于光栅投影法的茶叶嫩梢定位系统，能够有效识别嫩梢并获取其三维信息，为研究名优茶智能采摘机解决了茶叶嫩梢定位问题。

为了提高采摘机器人的定位速度,对机器人的机器视觉系统进行了改进,设计了一种基于聚类算法和视频对象提取技术的快速定位机器人。该机器人视频对象图像提取过程中,在完成图像进行滤波后,引入了Lab彩色空间聚类算法,有效地降低了图像的色彩数和噪声,实现了图像对象的量化处理,大大提高了果实定位和采摘的效率。为了验证设计的快速定位采摘机器人的可靠性,对机器人的采摘性能进行了测试,测试项目主要包括图像处理和果实定位。通过测试发现:快速定位机器人可以有效地实现图像聚类中心的提取,并对聚类中心进行编码,每次定位用时少、定位速度高且果实采摘的准确性累计概率较高,符合高精度、高效率果实采摘机器人的设计需求。

研究彩色图像目标定位精度问 题。对目前常见的图形图像中目标定位算法进行研究总结,分析了目标定位抽象流程,阐述了基于改进灰度模型的目标定位算法,提出了一种主动形状模型和HSV 颜色模型,以求取优化定位效果、降低算法的时间和空间复杂度为目标,设计了相关改进思路。将基于改进主动形状模型的彩色图像中目标定位算法应用到彩色视频 图像中,针对处理结果进行科学评价。通过实际的应用,再次验证了提出算法的可行性和高效性。

【Objective】The purpose of the study was to create database of characteristics from maize leaf transmission images, analyze the rules of characteristics variation with maize varieties and the recognition results of different types of characteristics in order to provide a basis for further research of identifying maize varieties from leaf transmission image of different growth periods based on machine vision. 【Method】 Twenty-one common varieties of maize were selected as the research materials. The maize leaves at jointing stage, small bell stage, large bell stage and tasselling stage were collected. A total of 420 high quality transmission images of maize leaves were taken in lamp box. The software for characteristic extraction and recognition of maize leaves was designed and developed based on Matlab R2009a, which included image preprocessing module, characteristic extraction module, neural network recognition module and threshold selection module. The transmission images of maize leaves at jointing stage, small bell stage, large bell stage and tasselling stage were pre-processed by the software. Then 48 characteristics of color group, shape group and texture group were extracted from transmission images of maize leaf, and a total of 20 160 characteristic data. In order to study the rules of characteinristics variation with maize varieties, the coefficient of variation of 48 characteristics of leaf transmission image among different maize varieties were analyzed. In order to search the important characteristics with strong ability of identifying maize varieties from transmission images of leaves, the Artificial Neural Network was built and the recognition rate of single characteristics from different time were analyzed. In order to study the recognition results, the recognition rates of the three groups of characteristics and the group combinations of characteristics from different time were further analyzed. 【Result】 The results in 4 stages indicated that there were significant differences in the coefficient of variation of 3 groups of characteristics among different maize varieties. The differences were stable with the growth of maize. The coefficient of variation of color group was the highest, then the texture group and the third was the shape group. The results at 4 stages also indicated that there were significant differences among the recognition rates of 48 single characteristics. The recognition rates were between 9.52% and 29.33%. According to the recognition rates, the important characteristics were in the following order: the standard deviation of R, the minor axis length, the standard deviation of H, the diameter, the average value of H, the standard deviation of V, the standard deviation of B, the invariant moment 6, the eccentricity, the average value of S, the external convex polygon area, the average value of B, the smoothness, the kurtosis of V, and the standard deviation of S. Those average recognition rates was over 18%. The average recognition rate of the color group was the highest which was 86.76%, then the texture group which was 78.05%, and the third was the shape group which was 68.67%. The stability of recognition rates of the texture group was the highest, then the color group and the third was the shape group. The average recognition rate of combinations of shape group and color group was the highest which was 92.29%, then the combinations of color group and texture group which was 90.29%, and the third was the combinations of shape group and texture group which was 87.43%. At jointing stage, there was no obvious difference among the recognition rates of combinations of 3 groups. At small bell stage, the recognition rate of the combinations of color group and texture group rose sharply, while the recognition rate of the combinations of shape group and color group fell slightly, and the recognition rate of the combinations of shape group and texture group showed a marked decline, which still remained above 82%. There was no significant difference between recognition rates of the shape group and texture group and the color group and texture group at jointing stage, large bell stage and tasselling stage. And the recognition rate of combinations of the shape group and color group was always the highest. 【Conclusion】 This study has provided rich primary data on characteristics from maize leaf transmission images, and has provided a quantitative basis and a new method for further research of identifying maize varieties in different growth periods, which will play a good reference role in crop varieties identification.

【目的】建立玉米品种的叶片透射图像特征数据库，研究特征随品种的变化规律，分析各类特征的识别效果，为进一步研究玉米生长期间的机器视觉品种识别提供依据。【方法】以生产中推广的21个常规玉米品种为供试材料，分别采集拔节期、小喇叭口期、大喇叭口期、抽雄开花期4个生育时期的玉米叶片。在灯箱内，采集每一叶片的高画质透射图像，共计420张。基于Matlab R2009a开发了“玉米叶片特征提取与识别软件”，包括图像预处理、特征提取、神经网络识别和阈值选取4个功能模块。依据开发的特征识别平台，对玉米叶片透射图像进行图像预处理和特征提取。提取形态类、颜色类和纹理类共计48个特征，特征数据量共计20 160条。分析48个特征品种间的变异系数，研究玉米叶片透射图像特征随品种的变化规律。建立BP神经网络模型进行综合识别，分析不同时期单特征的识别效果，寻找玉米叶片透射图像中品种区分能力较强的重要特征。进一步分析不同时期3大类特征及其组合的识别效果。【结果】在玉米的4个生育时期，叶片透射图像3类特征品种间的变异系数差异比较明显，颜色类特征变异系数最大，其次是纹理类特征变异系数，形态类特征变异系数最小，并且这种差异随着玉米的生长十分稳定。在玉米的4个生育时期，叶片透射图像48个特征的品种识别率差异比较明显，为9.52%—29.33%。R分量的标准差、短轴长、H分量的标准差、等面圆直径、H分量的平均值、V分量的标准差、B分量的标准差、不变矩6、椭圆度、S分量的平均值、外接凸多边形面积、B分量的平均值、平滑度、S分量的峰度、S分量的标准差的识别率较高，平均识别率在18%以上。单类特征中，颜色类特征识别率最高，平均86.76%；纹理类特征次之，平均为78.05%；形态类特征最低，平均为68.67%。颜色类特征和纹理类特征识别的稳定性较高，纹理类特征识别效果更稳定一些。组合特征中，形态+颜色特征组合识别率最高，平均识别率为92.29%；颜色+纹理类特征组合次之，平均为90.29%；形态+纹理类特征组合识别率最低，但平均识别率也达到了87.43%。在拔节期，3类特征组合的识别率无明显差异，且都在91%以上。在小喇叭口期，颜色+纹理特征组合识别率大幅上升，形态+颜色特征组合识别率小幅下降，形态+纹理特征组合识别率下降较大，但是仍然维持在82%以上。在其他3个时期，形态+纹理特征组合和颜色+纹理特征组合识别率差别不大，并且形态+颜色特征组合识别率相对较高。【结论】研究结果为玉米叶片透射图像特征的研究与应用提供了比较系统全面的数据，为生长期间玉米品种的识别提供了新的方法和量化依据，也必将在其他作物的识别方面发挥很好的借鉴作用。

Digital image analysis was used to develop a pattern recognition algorithm to classify individual kernels of seven Chinese spring wheat cultivars grown at 4 locations. Totally, 20 morphological parameters and 12 color parameters were extracted. Three hundred kernels per sample were used as the training data set to develop identification model, and another 200 kernels were used as the test set. For the test set, the classification accuracy of wheat cultivars was 100% in each growing location. Except for Xinkehan 9 with 98.3%, the correct discrimination of the training set of collective samples is 100% for wheat cultivar. For the test set, the correct discrimination of Longmai 26 and Qingchun 566 were 97.5% and 95.0%, the others is 100%. For the origin of wheat grains, the classification of Gansu, Ningxia, Xinjiang and Heilongjiang were 88.6%, 92.9%, 72.9% and 95.7%, respectively. The results show that it is feasible to identify and classify wheat cultivar (grains) using digital image analysis.

基于数字图像分析，利用小麦籽粒的20个形态特征和12个颜色特征对来自中国4个地点7个春小麦品种共28个样本进行分类和识别。对于不同品种和地区的样本，分别利用逐步判别分析，选取显著性较大的特征参量，建立各地区和品种的贝叶斯分类器模型。结果表明，对各地区品种识别的正确回判率和测试集的正确识别率均达到100%。将各样本按品种合并，再对合并后的样本进行品种识别，除了新克旱 9号的回判率为98.3%外，其它品种的回判率均为100%。测试集中，龙麦26和青春566正确识别率分别为97.5%和95.0%，其它品种均为100%。品种来源地识别也能达到较高的水平，甘肃、宁夏、新疆和黑龙江的正确识别率分别为88.6%、92.9%、72.9%和95.7%。说明利用籽粒图像对小麦品种进行识别高效可行。

【Objective】 The recognition and diagnosis methods of main maize leaf diseases using machine vision were studied in this paper. 【Method】 The diseases pictures of different varieties or periods were taken in fields, methods of threshold segmentation based on hue, iteration binarization, image morphological operation and contour extraction were adopted for image processing and image segmentation, then the texture, color and shape features were extracted. Genetic algorithm was used to get approximate features. Finally Fisher discrimination analysis was applied to recognize main maize leaf diseases. 【Result】 In the research, 28 characters including energy, informationization measure, fractal dimension, hue, cb, color moment, disease spot area, rotundity, figure factor, and others were extracted, and four approximate features were selected from 28 primordial features. The results indicated that the precision of the three kinds of maize disease recognition was higher than 90%. 【Conclusion】 Disease image obtained in fields were recognized by the application of digital image processing technology , analysis of image texture, color and figure characters, genetic algorithm and Fisher discrimination analysis. It has provided a technical support for the automatic recognition of crop diseases and insets with disease image obtained in fields.

Image recognition with fruit color character in natural scene plays an important role for efficient and real-time recognition. Eighty images were recognized with an R/B segment threshold of 1.375. The rates of detection, successful recognition, false recognition and missing recognition were 98.3%, 86.1%, 12.4% and 13.9% respectively. Furthermore, the above rates were 90.3%、84.9%、6.0% and 15.1% with an additional V segment threshold of 0.45. The comparison results showed that although the rate of successful recognition decreases slightly, the rate of false recognition fell remarkably. A series of test were executed under direct sunlight with strong, medium and weak illumination, and backlighting with strong, medium and weak illumination. The results showed that the rate of detection and successful recognition were the highest of 91.7% and 87.7%, and the rate of miss recognition was the lowest of 12.3% under direct sunlight and medium illumination. The rates of successful recognition in various conditions were different, but not significant.

在自然场景下，利用果实自身颜色特性进行图像识别对于提高识别效率和实时性具有重要作用。该文采用1.375作为R/B的分割阈值，对80幅成熟期苹果图像进行识别，其识出率、识别成功率、误识率和漏识率分别为98.3%、86.1%、12.4%、13.9%；在此基础上，加入0.45作为V的分割阈值，其识出率、识别成功率、误识率和漏识率分别为90.3%、84.9%、6.0%、15.1%，由此可见，基于R/B值和V值的混合颜色空间识别方法，在小幅降低识别成功率的情况下大幅降低误识率。通过进行顺光弱光照条件、顺光中光照条件、顺光高光照条件、逆光弱光照条件、逆光中光照条件、逆光高光照条件下的识别测试，结果表明顺光中等光照条件下的识出率和识别成功率最高，分别为91.7%和87.7%，漏识率最低为12.3%；不同光照条件下的识别成功率存在差异，但不明显。

在数学形态学的理论基础之上,运用8种结构元素模板对图像进行击中匹配迭代运算,不断剥落图像边缘点,从而得到图像的中心骨架,达到细化的目的.该细化算法的连通性、单像素宽度性、抗噪声性都非常良好,而且处理速度很快,非常适合于有大量图像处理运算的反求工程.

图像的细化是提取曲线特征,矢量化、识别等后续图像处理的基础,而单一的算法往往不能得到理想的细化效果。本文在研究各种已有的细化算法的基础上,从中选出两种不同的数学形态学算法共同处理同一图像,可以实现较好的细化效果:即先对图像采用基于5×5邻域模板的数学形态学细化算法进行细化处理,再对处理后的图像采用基于结构元素对的数学形态学细化方法处理。通过编程和各种方法相比较,证明采用这种方式可以实现无腐蚀,连通性好,无毛刺,且为八邻连接的细化效果,为后续的图像处理打下良好的基础。

对目标图象进行细化处理，提取其骨架，在目标检测、图象编码以及光学字符识别等计算机视觉、图象处理与模式识别领域都有着广泛的应用．在定义了数学形态学图象轮廓和骨架的基础上，提出了基于两组结构元素模板的数学形态学细化新算法，这两组结构元素模板分别用来去掉目标的西北、东北、东南、西南方向四个角上的点和北、东、南、西四个方向上的点．用该算法对实际的字符进行细化，获得了令人满意的结果．

研究了基于HSV颜色空间的粗糙集直方图阈值彩色图像分割方法.根据HSV空间的特性,定义了HSV空间颜色距离计算公式,对HSV三分量计算粗糙集直方图和彩色图像进行分割.实验结果表明该方法较好地保留了图像的颜色信息,具有较高的通用性,能够获得满意的分割效果.