【作者】 李志臣；

【导师】 姬长英；

【作者基本信息】 南京农业大学 ， 农业机械化工程， 2007， 博士

【摘要】 化学除草剂的使用方式普遍为粗放式的大面积喷洒,一方面容易造成人身伤害,另一方面因为喷药时不是精确对准目标物,往往也喷在了无目标的土壤上。这种喷药方式不仅浪费了农药,还造成了生态环境危害,增加了作物的农药残留量,随着农药的大量使用,其残留在农作物和土壤中的农药对人身和环境的危害也与日俱增,这跟现代农业要求精准高效、绿色环保的要求是背道而驰的。为了减少成本和保护环境,人们开展了除草剂的精确喷药技术的研究,希望替代传统的大面积喷药技术,以降低人的劳动强度,提高药效和保护农田环境。杂草在田间的分布是随机的,不均匀的,而且具有簇生性。正是这种杂草分布的不连续性,为精确施药技术的可行性提供了依据。因为条播作物的行距基本恒定,条播作物都成行排列,从而有利于基于机器视觉的杂草识别,在总结国内外相关研究的基础上,本文以条播作物-小麦为主要研究对象,设计和开发了基于机器视觉的杂草对准喷药控制系统。本文的主要研究内容如下:1.研究了基于颜色特征的图像背景分割,对十三个分割因子就分割质量和分割所需的时间进行了对比研究。通过研究发现过绿的2G-R-B是最佳的分割因子。2.自动取阈值法分割图像能够大大提高图像处理的实时性,本文设计了最大类间方差和最小错误率相结合的自动取阈值算法,并与最大方差自动取阈值法和最小错误率自动取阈值法进行了对比研究,实验证明该算法好于其它两个算法。3.针对杂草类植物的形态结构具有自相似特征,研究了杂草整体形态和杂草叶的分形特征,设计了计算分形维数的算法和程序,对杂草分形维数的分析显示,运用杂草的分形维数能够有效地识别杂草。4.本文分析和讨论了基于灰度共生矩阵的纹理统计值的计算方法以及在不同光照条件下的杂草和麦苗的纹理统计值,结果显示杂草和麦苗的纹理统计值对光照不是很敏感,能够运用杂草和麦苗的纹理统计值识别杂草。5.针对条播作物的行距基本恒定,条播作物都成行排列的位置特征,本文着重研究了作物中心行的提取方法,设计了改进的像素直方图法提取作物中心行的算法。通过分段提取作物中心行,有效地提高了算法的田间适应度。还研究了另外三种提取作物中心行的方法。6.针对人工神经网络在模式识别上的优势,设计了三层BP人工神经网络,并运用三层BP人工神经网络进行了杂草识别的实验。结果显示,无论是基于纹理还是基于分形,运用人工神经网络都能够很好地识别杂草。7.设计和开发了基于机器视觉的麦田除草剂对准喷药控制系统,该系统的视觉子系统首先用过绿的2G-R-B进行植物与土壤背景的分割,其次用改进像素直方图法识别作物中心行,用阈值法识别行间杂草区域,用BP神经网络基于分形和纹理的共同特征识别行内杂草区域,设计并开发了计算机与PLC的通信程序,设计了PLC的运行程序。设计并装配了喷药执行系统。8.在校园里的一块水泥地上的模拟麦田进行了除草剂的对准杂草喷药控制的实验,实验结果表明喷头能够有选择地对准杂草区域喷药。更多还原

【Abstract】 The usage method of chemistry herbicide is wide-spread to spray, spraying isn’t to aim at a target thing accurately, but usually spraying at the soil without weed, so plenty of herbicide is wasted , resulting in not only the Human body hurt easily, but also the ecosystem environment crisis and increasing agricultural chemicals residue. Along with a great deal of usage of pesticide and herbicide, the hurt to the human body and the environment increase with each passing day. It is opposite to the modern agriculture which require precision, efficiency and Environment friendliness. For reducing cost and protecting environmental, the research of spraying precisely is done by more and more people, hoping it can take the place of the traditional spraying method, so the labor condition and the efficiency of herbicide can be improved and the pollution of environment can be avoided.The distributing of weed in the field is random and asymmetry. The discontinuity of weed distributing provide a basis for the precise-spraying. Because the constant row spacing is good for weed identification, the spraying control system aiming at weed in row wheat based on machine vision was designed and developed on the basis of research on abroad and domestic. The contents of the study could be briefly summarized as follows:1.Image segmentation based on color characteristic is researched. The segmented result and consuming time about 13 factors of color were researched contrastively. The research result was the exceed green 2G-R-B is the best segmented factor.2.Image segmentation based on taking threshold automatically can process the image in time. The algorithm of combining maximum variance and minimum error is designed. The contrast research certificate the algorithm is better to the other two algorithm.3.Aiming at the alike characteristic of weed structure, the weed and weed leaf fractal characteristic was researched and algorithm of calculating fractal dimension is designed. The analysis of weed fractal dimension verified it can be used for weed identification effectively.4.The calculating method of statistical texture value based on gray co-occurrence and the texture value of weed and wheat under different sunshine were analysised. The texture value is not sensitive to the sunshine and it can be used for weed identification. 5.Aiming at the position characteristic of constant spacing, the method of obtaining the center line of crop rows is researched. The improved pixel lateral histogram algorithm for extracting the center of crop row is designed. The infield adaptability of the algorithm was enhanced by subsection computation. Other three method of extracting the center of crop row is also studied.6. Aiming at advantage of pattern recognition of ANN, three layer BP ANN is designed for the experiment of weed identification. The experimental result is that the ANN cold be used for weed identification based on texture value or/and fractal dimension.7.The spraying control system aiming at weed based on machine vision was designed and developed. The index of 2G-R-B is used for segmentation and the method of improved pixel lateral histogram is used for extracting the center of crop row,the method of threthold is also used for identificating between-row weed and BP ANN based on fratal and texture is used for identificating in-row weed. PLC correspondence and performing procedure is desi -gned and developed. Spraying performing system is designed and assembled.8. The experiment was done on a plot modifying wheat field, the result verified that the control system is able to spray selectively.更多还原