【作者】 刘芳；

【导师】 倪军；

【作者基本信息】 南京农业大学 ， 农业推广硕士（专业学位）， 2017， 硕士

【摘要】 稻麦是世界主要粮食作物,快速精准地获取稻麦长势信息对于管理者做出生产指导、病虫害预防和施肥调控等决策极其重要。株高、覆盖度、叶片叶绿素相对含量、叶面积指数是稻麦基本的长势指标,长期以来主要依靠人工经验观察、手工测量获取,这种方法简单易行,但主观性强;室内化学定量分析,需要破坏性取样,时效性较差;近年来,新兴的光谱检测技术,因其测量方法便捷、高效无损,逐渐成为支持地物识别的一项关键性技术,但仍然面对着测量费用昂贵的问题,不利于推广。随着图像传感器和高速处理器的飞跃发展与普及,数字图像处理技术逐渐渗入到农业信息获取领域中,主要用于作物产量预测,病虫害检测,杂草识别,营养监测,农作物质量检测,作物形态测量等方面。本研究采用数字图像处理技术研发了稻麦长势数字图像分析系统,该系统可以实现稻麦图像采集、处理、特征提取以及农学参数解译等功能,可检测稻麦的株高、覆盖度、叶片叶绿素相对含量、叶面积指数状况。主要研究内容及成果如下:1.稻麦长势数字图像采集与处理;根据应用需求选择CMOS图像采集模块,同时对比分析现行通信方式和智能终端设备性能,集成稻麦长势数字图像获取CMOS-以太网-PC平台硬件模块,可以采集冠层俯视图、株系侧视图图像。对比分析了各图像处理算法处理后的结果,选定图像处理方案:对比图像增强(各通道灰度直方图)结果,根据双峰原则,选定俯视图的2*G-R-B通道与侧视图的G-R通道;对比图像分割(最大类间方差法、迭代法、最大摘法、sobel边缘检测法、laplacian边缘检测法、canny边缘检测法)效果,根据有监督评价,选定最大类间方差法;对比图像复原(中值滤波、均值滤波、维纳滤波)结果,根据峰值信噪比,选定中值滤波法。提取颜色特征(平均颜色特征值、比值颜色特征值、差分颜色特征值、归一化差分颜色特征值、归一化颜色特征值、综合颜色特征值),纹理特征值,形态特征值。为下一步建模提供数据依据。2.稻麦长势数字图像分析系统的模型构建;使用Photoshop软件、SPAD-502叶绿素仪、LAI-2200C冠层分析仪分别测定冠层的覆盖度、叶片叶绿素相对含量、叶面积指数,使用卷尺测定株高。将提取的图像特征值与农学参数进行相关性分析。利用相关性较高的图像特征值建立回归模型。根据相关系数(Correlation coefficient,r)、决定系数(coefficient of determination,R2)、均方根误差(root mean square error,RMSE)和相对误差(relative tolerance,RE)判断分析各图像特征值与稻麦农学参数的相关程度以及关系模型预测精度。最终建立图像提取株高与实测株高的估测模型y=1.057x-2.769,R2为0.856,RMSE为5.794;对模型进行验证得到预测株高与实测株高的R2为0.706,RMSE为0.374;建立图像提取的覆盖度与利用Photoshop软件人机交互方式获取的覆盖度的估测模型y=0.911x+0.030,R2为0.753,RMSE为0.054;对模型进行验证得到预测覆盖度与实测覆盖度的R2为0.939,RMSE为0.031;图像颜色特征值g与叶绿素的相关系数为0.916,建立估测模型y=34.467x+27.035,R2为0.839,RMSE为1.541;对模型进行验证得到的预测叶片叶绿素相对含量值与实测SPAD值的R2为0.697,RMSE为2.909;图像颜色特征值G-B与叶面积指数的相关系数为0.803,建立估测模型y=0.279e0.038x,R2为0.833,RMSE为0.64;对模型进行验证得到预测叶面积指数与实测叶面积指数的R2为0.790,RMSE为0.158。3.稻麦长势数字图像分析系统的应用软件研发;利用Matlab软件的GUI功能设计并实现了基于Windows平台的稻麦长势数字图像分析c/s系统。主要功能包括图像读取(JPEG文件)以及显示,图像处理(多种图像通道提取、六种图像分割、三种图像复原)以及结果显示,图像特征提取(颜色特征值、纹理特征值、形态特征值)以及数据显示,图像解译(稻麦株高、覆盖度、叶片叶绿素相对含量、叶面积指数)以及结果显示,图像数据清空复位。结果表明,所设计的稻麦长势数字图像分析系统操作便捷,交互特性良好。成功探索了稻麦株高、覆盖度、叶片叶绿素相对含量、叶面积指数的估测方法,建立的估测模型具有实用性。更多还原

【Abstract】 Rice and wheat are the major food crops in the world,instant and accurate access to information of rice and wheat growing is critical.It is convenient for the managers to make regulatory decisions on the production,pest control,fertilization and so on in timely.In the detection of the growth status of rice and wheat,plant height,coverage,relative chlorophyll contents of leaves and leaf area index are the basic growth indexes of rice and wheat.Their acquisition rely on the artificial experience observation,manual measurement,the method is simple and feasible,but it is subjective;quantitative analysis of indoor chemical,require destructive sampling and poor timeliness;in recent years,spectrum detection as new technology has gradually become a key technical support for object recognition,because of its convenient and efficient nondestructive measurement method,but it still faces the problem of high measuring cost,It is not conducive to promoting.With the rapid development of image sensor and the high-speed processor,the popularity of digital image processing technology is gradually infiltrated into the agricultural information acquisition in the field,mainly used for the prediction of crop yield,insect pest detection,weed identification,crop nutrition monitoring,quality inspection,measurement and other aspects of plant morphology.This study adopts digital image processing technology research and development of rice and wheat growing digital image analysis system,the system can realize image acquisition and processing of rice and wheat image,feature extraction and agronomic parameters interpretation and other functions,can detect in rice and wheat plant height,coverage,relative chlorophyll contents of leaves,leaf area index.The main research contents and results are as follows:1.Digital image acquisition and processing of rice and wheat growth;Select the CMOS image acquisition module according to the application demand,and comparative analysis of current communication mode and intelligent terminal equipment performance,to integrate rice and wheat growing digital image acquisition CMOS-Ethernet-PC platform hardware modole,it can collect the canopy top view,side view image.Comparative and analysis the results of each image processing algorithm,to selecte image processing scheme:contrast image enhancement(each channel histogram)results,according to the Shuangfeng principles,selecte 2*G-R-B channel of top view image and G-R channel of side view image;contrast image segmentation(Otsu method,iterative method,maximum entropy method,Sobel edge detection method,Laplacian edge detection method and Canny edge detection method),according to the supervised evaluation,selecte Otsu method;contrast image restoration(median filtering,mean filtering,Wiener filtering)results,according to the peak signal-to-noise ratio,selecte median filtering method.Extract color feature value(average color feature value,the ratio of color difference color characteristic value,characteristic value,normalized difference color feature value,normalized color characteristic value,comprehensive color feature value),extract texture feature value,extract shape feature value.Provide data basis for the next modeling.2.Model construction of digital image analysis system for rice and wheat growth;The canopy coverage,relative chlorophyll contents of leaves and leaf area index were measured by Photoshop software,SPAD-502 chlorophyll analyzer and LAI-2200C canopy analyzer respectively.the plant height were measured by tape.Analyze the correlation between the extracted image features and agronomic parameters.The regression model is established by the high correlation image feature values.According to the correlation coefficient,the coefficient of determination,root mean square error and relative error,determine the degree of correlation with the image characteristic value of rice and wheat agronomic parameters and the prediction accuracy of the model.Finally,the estimated model of image plant height and real plant height was y=1.057x-2.769,R2 was 0.856,RMSE was 5.794,and the R2 of prediction plant height and actual plant height was 0.706,RMSE was 0.374;The estimated model of image coverage and the coverage obtained by Photoshop software was y=0.911x+0.030,R2 was 0.753,RMSE was 0.054,and the R2 of prediction coverage and the actual coverage was 0.939,RMSE was 0.031;The correlation coefficient of the color feature g and relative chlorophyll contents of leaves was 0.916,and the estimated model was y=34.467x+27.035,R2 was 0.839,RMSE was 1.541;The R2 of prediction relative chlorophyll contents of leaves and the actual SPAD value was 0.697,RMSE was 2.909;The correlation coefficient of the color feature G-B and leaf area index measured by LAI-2200C was 0.803,and the estimated model was y=0.279e0.038x,R2 was 0.833,RMSE was 0.64;The R2 of prediction leaf area index and the actual leaf area index was 0.790,RMSE was 0.158;3.Research and development of the application software of rice and wheat growth digital image analysis system;Based on the Windows platform,the c/s image analysis system of rice and wheat growth was designed and realized by using the GUI function of Matlab software.The main features include image reading(JPEG file)and display,image processing(image channel extraction,six kinds of image segmentation,three kinds of image restoration)and the results display,the image feature extraction(color feature value,texture feature value and morphological characteristics value)and data display,image interpretation(rice and wheat plant height,coverage degree,relative chlorophyll contents of leaves,leaf area index)of the image data and display the result,empty reset.The results show that the design of digital image analysis system for rice and wheat growth is easy to operate and has good interaction.The estimation methods of plant height,coverage,relative chlorophyll contents of leaves and leaf area index were successfully explored.The established estimation model is practical.更多还原