一种基于图像特征值算法的叶面积测定方法

提出一种基于图像特征值算法的叶面积测定简化方法。应用扫描图像RGB三原色灰度值分离理论,根据植物叶片扫描图像像素点的分布特征,选用蓝色灰度值作为特征值,以扫描图像灰度中间值127作为叶面积图像与背景图像灰度值的判读指标,通过叶片像素点的分布比例计算叶片面积。将已知面积的矩形绿纸片分别随机裁剪成多个碎片,用本文方法测定碎片面积,并分别计算每个叶片的碎片面积之和进行系统精度验证,测定结果与标准面积的相对误差小于0.5%。采集60个水稻叶片分别采用本文方法和复印称重法测定叶片面积,对本文方法进行进一步验证,相关性分析结果表明,二者相关系数r=0.997 1,达极显著水平。本文方法具有较高测定精度,满足叶面积测定要求。     

基于RGB统计特征的农作物虫害图像感知系统设计

基于RGB统计特征,设计了一套农作物虫害图像感知系统,以实现对农作物生长状态的实时监测与分析。系统以视频服务器为核心,在连续视频采集周期内通过自相关函数的计算,筛选出消除外部随机干扰的基准图像,对相邻的基准图像进行差分运算,消除相似像素并滤波后统计其RGB特征值。最后,在视频服务器中用VC++开发了人机交互界面,移动客户端可以通过桌面远程控制实现与服务器数据同步。     

Modeling Dynamics of Leaf Color Based on RGB Value in Rice

This paper was to develop a model for simulating the leaf color changes in rice （Oryza sativa L.） based on RGB （red, green, and blue） values. Based on rice experiment data with different cultivars and nitrogen （N） rates, the time-course RGB values of each leaf on main stem were collected during the growth period in rice, and a model for simulating the dynamics of leaf color in rice was then developed using quantitative modeling technology. The results showed that the RGB values of leaf color gradually decreased from the initial values （light green） to the steady values （green） during the first stage, remained the steady values （green） during the second stage, then gradually increased to the final values （from green to yellow） during the third stage. The decreasing linear functions, constant functions and increasing linear functions were used to simulate the changes in RGB values of leaf color at the first, second and third stages with growing degree days （GDD）, respectively; two cultivar parameters, MatRGB （leaf color matrix） and AR （a vector composed of the ratio of the cumulative GDD of each stage during color change process of leaf n to that during leaf n drawn under adequate N status）, were introduced to quantify the genetic characters in RGB values of leaf color and in durations of different stages during leaf color change, respectively; FN （N impact factor） was used to quantify the effects of N levels on RGB values of leaf color and on durations of different stages during leaf color change; linear functions were applied to simulate the changes in leaf color along the leaf midvein direction during leaf development process. Validation of the models with the independent experiment dataset exhibited that the root mean square errors （RMSE） between the observed and simulated RGB values were among 8 to 13, the relative RMSE （RRMSE） were among 8 to 10%, the mean absolute differences （da） were among 3.85 to 6.90, and the ratio of da to the mean observation values （Clap） were among 3.04 to 4.90%. In addition, the leaf color model was used to render the leaf color change over growth progress using the technology of visualization, with a good performance on predicting dynamic changes in rice leaf color. These results would provide a technical support for further developing virtual plant during rice growth and development.     

Computer vision systems (CVS) for moisture content estimation in dehydrated shrimp

This paper presents a method based on computer vision systems (CVS) to estimate shrimp dehydration level by analyzing color during drying process. Since the most commonly used color space in food industry is L*a*b, transformation of RGB digital images to L*a*b units was carried out using direct two steps model with γ factor. Experimental data obtained from images captured at different drying temperatures (100–130 °C) and several time intervals (15–180 min) were analyzed with a complete randomized block design (CRBD), and the means were compared with Duncan's multi-range test. Multiple linear regression (MLR) and artificial neural networks (ANN) were applied for correlating the color features to moisture content of dried shrimp determined chemically. Results obtained with these two models lead to 0.80 and 0.86 correlation coefficients in MLR and ANN models, respectively. While there is no statistical difference at p < 0.05 between the two modeling approaches, both approaches indicate successful prediction of shrimp dehydration with high correlation to those found by the more expensive and intrusive chemical method. The automated vision based system, therefore, has the advantage over conventional subjective methods and instrumental ones for being objective, fast, non-invasive, inexpensive and precise.     

基于SURF算法的绿色作物特征提取与图像匹配方法

由于田间环境的复杂性,绿色作物特征提取与匹配仍然是基于双目视觉技术农田作物三维信息获取急需解决的关键技术之一。该文首先在RGB空间进行图像分割滤波处理。然后,采用SURF算法旋转不变性分两步获取绿色作物特征点对:第一步采用Hessian矩阵检测作物特征点,运用非极大值抑制法和插值运算寻找、定位极值点;第二步确定特征点主方向,采用描述算子进行特征点提取。最后,运用最近距离比次近距离法进行特征点匹配,并采用全约束条件滤除错误的匹配点对。同时将SURF和SIFT法进行对比分析,通过对不同光照、土壤的田间条件下芥蓝、芹菜、白菜13组图像进行试验,结果表明采用SUFR和SIFT法绿色作物特征提取率均值分别为1.2%、3.3%,双目视觉系统左、右作物图像特征正确匹配率的均值分别为94.8%、92.4%,时间消耗均值分别为4.6s、4.8s。采用SURF优越于采用SIFT法,这为进一步进行农业机械3D视觉导航或基于无线传感器网络的田间作物在线三维信息准确获取提供可借鉴思路和方法。     

基于kmeans的雪茄烟晾制阶段识别研究与应用

【目的】为对雪茄烟晾制阶段进行精准分类,提升晾制管理精细化。【方法】采集晾制过程的重量及图像,并获取图片RGB、HSV颜色空间的值,及纹理特征能量(Asm)、对比度(Con)、熵(Eng)、逆方差(Idm)进行Kmeans聚类分析,并依据聚类结果对特征值进行一元线性回归直线方程的求解,并提取直线的斜率(K)与线性回归相关系数(R2)。【结果】晾制过程聚类的阶段与晾制过程连续性一致,各阶段起始与终止日期的烟叶图片具有显著差异,聚类过程选取的输入特征值有效。【结论】通过雪茄烟晾制过程中的失水率、RGB、HSV颜色空间、纹理特征对于判断雪茄烟晾制过程的阶段、指导雪茄烟晾制具有实用性。     

基于无人机可见光图像的荒漠草地覆盖度估算

及时准确监测草地植被覆盖度,对草地资源的可持续利用及生态系统的恢复与重建具有重要意义.本研究以荒漠草地植被为研究对象,采用监督分类与植被指数直方图相结合的阈值法,分析了6种RGB植被指数对荒漠草地的识别效果.研究结果表明:归一化绿红差异指数(normalized green-red difference index,N...     

基于无人机可见光图像估算甘肃鼢鼠对马铃薯的为害量

为估算农田害鼠对作物的为害损失量,使用无人机拍摄甘肃鼢鼠Eospalax cansus为害的22块马铃薯样地正射影像图,首先目视解译标定各样地为害区域,计算为害率,并据此划分各样地鼠害为害等级;随后运用基于规则的特征提取法和监督分类法（支持向量机分类法和神经网络分类法）对各样地裸地和植被进行分类,结合对照区裸地率计算各样地的鼠害为害裸地率;通过构建鼠害为害裸地率与马铃薯产量的线性关系模型来评估不同分类法获得的鼠害为害裸地率的精确性;用拟合度最好的线性关系模型估算无鼠害及当前鼠害水平下的马铃薯产量,最终计算全部样地鼠害造成的马铃薯损失量。结果表明,基于规则的特征提取法、支持向量机分类法和神经网络分类法的地物分类精度分别为71.46%、99.33%和98.84%,3种分类方法获得的样地鼠害为害裸地率与马铃薯实际产量均呈显著线性相关,但神经网络分类获得的结果拟合度最好,R²为0.558。利用该方法估算的甘肃鼢鼠造成的马铃薯产量损失量为7 032.75 kg/hm²。     

不同植被指数和无人机航高对草地盖度估测精度的影响

搭载高清数码相机的无人机在草地资源调查等方面具有成本低廉、机动性高、观察范围大等突出优势,拥有广阔的发展前景。本研究使用小型无人机(unmanned aerial vehicle, UAV)、手机相机等设备获取草地盖度数据,选用植被因子指数(vegetative index, VEG)、超绿指数(excess gree...     

基于机器视觉的冬小麦叶片形态测量软件开发

随着图像处理与识别技术的快速发展,作物表型识别技术日趋成熟。为实现不同品种、不同生育期冬小麦叶片面积和面积系数的精准快速测定,依托VB.net和OpenCV在.NET平台下的图像处理封装库,研发了基于机器视觉的冬小麦叶片形态测量算法并设计开发了软件,软件可实现数字图片的畸变校准并可以同时测量多个叶片长、宽和面积。为验证软件测定效果,选取冬小麦绿色展开叶100 片,通过与人工测量的叶片长宽、WinDIAS叶面积分析系统测量的叶面积结果对比,分析图像识别方法的准确性和稳定性。结果表明,图像识别法与人工和WinDIAS测量的冬小麦叶片长、宽和面积的相关系数均≥0.975,归一化均方根误差均≤0.10%;针对数字照片畸变校准功能进行测试,对叶片水平（垂直）缩放50%且垂直（水平）斜切30°的图像校准后,其测量结果与原始图像测量结果的最大相对误差仅为2%。说明基于机器视觉的冬小麦叶片形态识别方法,可对多种畸变图像进行准确的几何校准,可作为一种可同时准确测定多个叶片面积和长宽的新方法,在农业科学测量、农情信息业务、农业气象观测业务等领域推广应用。