基于图像分析技术的小麦群体农学参数获取与群体质量评价研究

发布时间：2018-01-07 04:16

本文关键词：基于图像分析技术的小麦群体农学参数获取与群体质量评价研究　出处：《扬州大学》2016年博士论文　论文类型：学位论文

【摘要】：目前,随着现代信息技术与农业产业的深度融合,农业生产将变得更加智能化,这将是我国现代农业发展的必然趋势。本文提出的基于图像分析技术的小麦群体农学参数智能获取与群体质量评价研究正是在这种现代农业发展的背景下展开,探求一套可以实现小麦生产智能化和管理高效化的新方法。研究以小麦生育进程为主线,探明了小麦苗期、越冬期、拔节期、孕穗期和成熟期主要农学参数的测算方法,并建立了小麦群体质量的评价模型,完成了小麦主要农学参数智能获取和群体质量智能化评价系统。研究结果可以为小麦物联网中的智能监控系统提供技术支持和理论依据,亦可为开发基于移动终端的智能田间测量和评价软件提供参考。主要研究结论如下：(1)构建了大田环境下苗期麦苗智能计数的方法。这部分内容建立了基于图像分析技术的野外环境下的麦苗智能计数方法,探明了大田环境下麦苗计数的原理,并验证计数方法在不同密度和品种条件下的适应性。研究选取5个不同株型品种和5种不同密度的小麦苗期图像作为研究对象,利用数码相机垂直获取图像,并利用超绿特征值(ExG)将小麦从背景中分离。分析了不同重叠麦苗区域的特征参数,建立了一种基于链码的骨架优化方法,并利用新骨架特征值提出了重叠区域麦苗计算公式。研究对5种不同播种密度的5个小麦品种共计250张图像进行计数测试,结果发现本研究提出的麦苗计数方法能够较好的对野外麦苗进行计数,平均计数准确率达89.94%,135×104株ha-1密度样本的计数准确率达到97.14%,在所有密度中最高,扬糯麦1号品种计数准确率达92.54%,在所有品种中最高。麦苗计数方法平均准确率89.94%,最高准确率达到99.21%,不同密度样本计数准确率之间达到了显著差异,而品种之间的差异没有达到显著水平(P0.05)。在田间苗数为120×104株ha-1至240×104株ha-1时本方法能够得到92%以上的准确率,说明本文设计的方法在麦苗计数上是可靠的,可为田间麦苗智能计数的研究提供理论依据,同样能移植到如水稻等禾本科作物的苗数智能计算上。(2)建立了越冬、拔节和孕穗期主要农学参数的估算模型。本研究拟利用图像分析技术建立小麦干物重、叶面积指数、茎蘖数和氮素积累量的估测模型,为这些农学参数的快速测量提供理论支持。通过不同的密度和氮肥施用量来构建具有不同农学参数的小麦群体,自群体越冬始期利用数码相机垂直获取冠层图像。研究通过超绿特征值(ExG)+自适应阈值分割(Ostu)的方法去除麦田耕地背景的影响,并用图像中小麦像素数占总像素数的比值表示盖度,另选取8种主要的图像特征算法提取图像的颜色和纹理特征,利用斯皮尔曼相关分析方法分析9种特征与不同时期农学参数的相关性。利用多元逐步线性回归方法建立基于图像盖度、颜色和纹理特征的农学参数估测模型。研究结果显示,本文提出的多元线性农学参数估测模型提高了单一参数模型的模拟精度,建立的4个模型对干物重、叶面积指数、茎蘖数和氮素积累量的预测效果较好,均具有较高的R2值,较低的RMSEP和REP。对于模型构建数据集的四个农学参数预测,R2值在0.77至0.91,REP值在15.46%至22.53%；验证数据集的R2值在0.72至0.85,REP值在17.31%至21.26%。本研究提出的多元农学参数估算模型能够较准确的估测出小麦群体的干物重、叶面积指数、茎蘖数和氮素积累量的值。(3)设计了成熟小麦穗数的智能化计算方法。为了实现不同播种方式下固定区域小麦穗数的智能计算,设计了一种利用图像分析技术实现大田麦穗快速计数的方法,着重分析了利用颜色特征和纹理特征分割麦穗的优缺点和粘连区域麦穗个数的计算方法。通过对撒播和条播多个样本图像进行计数实验,准确率分别为95.63%和97.07%。本研究结果说明,利用颜色特征和纹理特征均可以将麦穗从复杂的背景中提取出来,并可以通过形态学的腐蚀和膨胀以及孔洞填充算法得到麦穗的主要区域,然而利用颜色特征提取麦穗的速度远高于利用纹理特征提取。麦穗二值图像骨架的Harris角点能够较好的反映粘连区域的麦穗个数,Harris角点检测算法可以用于解决麦穗计数时粘连区域麦穗个数计算。本研究提出的麦穗计数方法在撒播小麦和条播小麦上的平均准确率分别为95.63%和97.07%。本研究提出的麦穗计数方法在不同品种上的平均高于95%,且麦穗计数结果在不同品种之间没有显著差异,说明该大田麦穗计数方法较为可靠,可以为大田麦穗的智能化计数提供有效的参考。(4)构建了基于BP神经网络的小麦群体质量评价模型。智能化地评价群体质量对于小麦智能化生产和快速制定栽培管理方案具有积极意义,完成群体质量评价模型的构建需要进行两部分工作：1)探明不同产量群体在不同生育期里表现的农学特征,明确高产群体在不同生育期的农学参数表现；2)构建不同生育期的小麦群体群体质量评价标准和评价模型。在第一部分研究中,试验选择扬糯麦1号为供试品种,采用二因素随机区组试验来构建不同结构的群体,设五个种植密度水平,四个氮肥施用量水平,重复两年。研究结果如下：1)探明了产量随种植密度变化的趋势和高产群体的种植密度范围；2)探明了越冬、拔节和孕穗期干物重、叶面积指数、茎蘖数、氮素积累量的变化对产量的影响和这4个农学指标在不同产量群体的区间；3)探明了产量随穗数的变化趋势和高产群体的穗数范围。这部分研究是群体质量评价的依据。依据前面所探明的苗数、干物重、叶面积指数、茎蘖数、氮素积累量、麦穗数与产量的关系,构建了基于这些农学参数的小麦群体质量评价模型。研究中通过K-means聚类算法对群体等级进行划分,以产量为标准划分各个时期的级别,同时基于这些农学参数的模拟值构建了用于评价小麦群体质量的BP神经网络模型,各个时期评价的依据分别为：1)苗期,以苗数为依据评价群体种植密度的合理性。2)越冬、孕穗和拔节期,以干物重、叶面积指数、茎蘖数和氮素积累量为依据,综合对这几个时期的群体质量进行评价。3)成熟期,以穗数为依据,判断群体穗数的合理性并对产量进行预测。研究解决了对小麦群体质量评价中各个农学参数与群体质量的非线性映射关系以及各个指标贡献率的问题。研究结果显示,研究中构建的群体质量评价模型在对苗期、越冬期、拔节期、孕穗期和成熟期的群体质量进行评价时,得到了较高的R2值和相对较低的RMSE值,说明模型可以用于评价小麦各个生育期的群体质量。该模型是后期开发群体质量评价系统和栽培决策的核心组成,亦可为其他作物群体质量评价提供一定的参考。(5)小麦群体农学参数测量与群体质量评价软件系统的构建。软件系统的构建是将此前提出算法的具体实践,是将小麦群体智能评价方法实用化的有效途径。本系统基于C/S的三层结构来开发,使用Microsoft Visual Studio 2013开发平台,MATLAB2014a图像处理和计算机视觉工具箱,SQL Server2013数据库完成开发。软件系统实现了麦苗计数和麦穗计数,越冬、拔节和孕穗期的茎蘖数、叶面积指数、干物重和氮素积累量的估测,以及各个生育时期群体质量的评价以及产量的预测。系统预留模型参数调节、品种群体质量标准添加接口和高产栽培方案添加接口,为后期在不同品种上的应用提供支持。本系统可为开展小麦田间智能感知和栽培决策的研究与应用提供一定的参考。
[Abstract]:At present, with the integration of modern information technology and agriculture industry depth, agricultural production will become more intelligent, this will be the inevitable trend of the development of modern agriculture in China. In this paper, the intelligent agriculture parameter of wheat based on image analysis technology for evaluation and Study on population quality is carried out in the context of the development of modern agriculture. To explore a new method of wheat production can achieve intelligent and efficient management. Research on wheat growth process as the main line, proved the wheat overwintering period, seedling stage, jointing stage, booting stage and calculation method of main agricultural maturity parameters, and establishes the evaluation model of wheat quality, complete the main agronomic parameters intelligent wheat population quality acquisition and intelligent evaluation system. The research results can provide technical support and theoretical basis for the intelligent monitoring system for wheat in the Internet of things, but also open Intelligent field measurement and evaluation software based on mobile terminal to provide the reference. The main conclusions are as follows: (1) the construction method of wheat field seedling intelligent counting environment. This part establishes the intelligent seedling counting method by image analysis technique based on field environment, proved the principle seedling counting under field environment, and validation of adaptive counting method in different density and variety conditions. The research chooses 5 different varieties and 5 different densities of wheat seedling image as the research object, the vertical image by digital camera, and the use of super green feature value (ExG) of wheat separated from the background. It analyzed the characteristic parameters of different overlapping region of wheat the establishment of a skeleton optimization method based on chain code, and use the new skeleton characteristic value calculation formula was proposed. The wheat overlapping area to study 5 kinds of different sowing density 5 wheat varieties with a total of 250 images were counting test, results showed that seedling counting method proposed in this study can count the number of wild barley, the average counting accuracy of 89.94%, 135 * 104 strains count HA-1 sample density accuracy rate reached 97.14%, the highest in all dimensions, Yang waxy wheat No. 1 varieties accurate count rate reached 92.54%, the highest in all varieties. Seedling counting method the average accuracy rate of 89.94%, the highest accuracy rate of 99.21% different density sample counting accuracy reached a significant difference, and the difference among the cultivars did not reach significant level (P0.05). The field seedling number is 120 * HA-1 to 240 * 104 strains 104 strains of HA-1 this method can obtain the accurate rate of more than 92%, shows that this design method is reliable in seedling counting, to provide a theoretical basis for research on the wheat field intelligent counting, can also be transplanted to such as The number of seedlings of rice and other cereal crops intelligent computing. (2) to establish the estimation model of main agronomic parameters of wintering, jointing and booting stage. This study intends to establish the wheat dry weight analysis technique using image, leaf area index, tiller number and nitrogen accumulation amount estimation model, to provide theoretical support for rapid measurement the agronomic parameters. The density and quantity of nitrogen fertilizer on different constructs with different agronomic parameters of wheat over wintering population groups, since the vertical access canopy image by digital camera. Through the study of super green feature value (ExG) + adaptive threshold segmentation (Ostu) effect method to remove the background and use of crop land. The image of wheat as the total number of pixels prime ratio of coverage, there were 8 main types of image feature extraction algorithm of color and texture features of the image, using the Spielman correlation analysis of 9 kinds of features and The correlation of agronomic parameters at the same time. By using multiple linear regression method based on image coverage estimation model of agronomic parameters of color and texture features. The results of the study showed that multiple linear agronomic parameters estimation model is proposed in this paper to improve the simulation accuracy of single parameter model, the 4 model of dry weight, leaf area index. Better prediction of tillers and nitrogen accumulation, have high R2 value, low RMSEP and REP. for four agronomic parameters model to construct the data set, R2 value is 0.77 to 0.91, REP value is 15.46% to 22.53%; the validation data set at R2 value of 0.72 to 0.85, REP value the model can accurately estimate the wheat dry weight estimation in multiple agronomic parameters of 17.31% to 21.26%. in this paper, leaf area index, tiller number and nitrogen accumulation value. (3) the design of mature spike number The intelligent calculation method. In order to realize the intelligent fixed area under Different Sowing Patterns in spike number calculation, design a kind of analysis method to realize fast counting of wheat field by the technology of image segmentation, analyzes the calculation method of wheat using color feature and texture feature and the advantages and disadvantages of the regional grain number. Adhesion by counting experiments to sow and drill a plurality of sample images, the accuracy rate were 95.63% and 97.07%. respectively. The results of this study show that the use of color features and texture features can be extracted from the wheat complex background, and by morphological erosion and dilation and hole filling algorithm to get the main grain area, however, the use of color feature extraction of wheat the rate is much higher than the use of texture feature extraction. The wheat two value image skeleton Harris corner can reflect a good regional wheat adhesion. The number of Harris corner detection algorithm can be used to solve the grain count number of wheat adhesion area calculation. Grain counting method proposed by this study in sowing wheat and wheat drill the average accuracy rate of respectively 95.63% and 97.07%. grain counting method proposed by this study in different species on average more than 95%, and the spike count results there was no significant difference among different varieties in wheat field, that the counting method is reliable, can provide effective reference for intelligent counting field wheat. (4) the construction of wheat quality evaluation group model based on BP neural network. The intelligent evaluation of population quality in wheat production and the rapid development of intelligent cultivation management scheme has a positive meaning, complete population quality evaluation model is constructed to two parts: 1) proved different yield group performance at different growth stages in agricultural Characteristics of high yield population in clear agronomic parameters in different growth stages; 2) construction of wheat population quality evaluation criteria and evaluation model in different stages. In the first part of the study, test Yang waxy wheat No. 1 as tested varieties, to construct different structure groups using two factor randomized block test five planting density levels, four nitrogen levels, repeated for two years. The results are as follows: 1) proved the range of planting density yield with planting density change trend and high yield population; 2) proved the wintering, jointing and booting stage of dry weight, leaf area index, tiller number, influence the accumulation of nitrogen on the yield and the 4 agronomy index in the interval of different yield groups; 3) proved the yield change trend with spike number and spike number of High-yielding Population range. This part of research is population quality evaluation according to the evidence. In front of the proven seedling number, dry weight, leaf area index, tiller number, nitrogen accumulation, the relationship between grain number and yield, construct the wheat population quality evaluation model based on the agronomic parameters. Research by K-means clustering algorithm is used to divide the population level to yield as the standard to divide each period at the same time, the simulation level, agronomy parameter values based on the BP neural network model for the evaluation of quality of wheat population was constructed based on the evaluation of each period are as follows: 1) at seedling stage, seedling number as the basis for evaluation of the rationality of.2 group planting density) wintering and jointing stage, booting stage, the dry weight, leaf area index. Tiller number and nitrogen accumulation based on the comprehensive quality evaluation of several groups during the mature period,.3) with spike number as the basis, judging the rationality of spike number and yield forecast research to solve the wheat group The problem of nonlinear mapping between the various agronomic parameters and population quality body quality evaluation and each index contribution rate. The results showed that the population quality evaluation model in the seedling stage, over wintering stage, jointing stage, booting stage and in population quality maturity evaluation, obtained a higher R2 value and the relatively low RMSE value, so the model can be used to evaluate the quality of wheat in different growth stages. The model group is the core of the late development group quality evaluation system and cultivation decision-making, but also provide some reference for other crop population quality evaluation. (5) the construction of wheat agronomic parameters measurement and population quality evaluation software system the software system is constructed. The specific practice of the previously proposed algorithm, is the effective way of practical evaluation method of wheat population intelligence. The system is based on the three layer structure of C/S The development, use Microsoft Visual Studio 2013 development platform, MATLAB2014a computer vision and image processing toolbox, SQL Server2013 database. The software system realizes counting and counting of winter wheat seedling, tiller number, jointing and booting stage, leaf area index, estimation of dry weight and nitrogen accumulation, and the assessment of population quality each growth period and yield prediction system. Reservation model parameter adjustment, breeds quality standard and cultivation scheme add add interface interface, provide support for the application in the late on different varieties. This system can provide some reference for the research and application of wheat field intelligent perception and decision to carry out the cultivation.

【学位授予单位】：扬州大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：S512.1

【相似文献】