归一化植被指数(NDVI)基于可见光的红色波段(630～680 nm)和近红外区(780～1 100 nm)的反射光谱进行计算, 被认为是作物营养与长势诊断的重要指标。 为了低成本、 快速、 无损的检测作物叶绿素含量, 计算植株的NDVI并呈现作物的NDVI分布情况, 并通过不同角度图像的分析, 监测作物营养分布与动态。 利用可见光和近红外波段双目成像技术获取图像, 在讨论可见光(RGB)和近红外(NIR)图像的匹配算法的基础上, 经图像分割与光照影响校正后, 针对不同测试角度建立了作物植被指数空间分布图, 并对其空间分布特征与影响因素进行了可视化分析。 试验利用可见光和近红外双目相机对51株玉米植株, 分别在90°, 54°和35°视角下同步采集RGB和NIR图像。 对RGB和NIR图像分别进行高斯滤波和拉普拉斯算子增强预处理后, 选取了SURF, SIFT和ORB共3种图像匹配算法, 并首先利用其进行RGB-NIR图像匹配对齐, 以匹配时间(Time), 峰值信噪比(PSNR), 信息熵(MI)和结构相似性(SSIM)4个参数作为匹配性能评价指标, 分别从时间、 准确性、 稳定性三个方面综合确定最优匹配方法。 其次, 研究玉米植株的分割方法包括超绿算法(ExG)和最大类间方差算法(OTSU), 分别实现图像中作物和背景的分离, 提取分割后的RGB图像R(Red), G(Green), B(Blue)分量和NIR图像分量。 基于HSI颜色模型, 提取I分量讨论了光照对图像的影响, 并利用多灰度级标准板建立了植株光谱反射率校正线性公式。 然后, 利用R(Red)和NIR图像分量计算图像中每个像素的NDVI值, 绘制作物植被指数的空间分布图, 从而对比分析了不同拍摄角度下光谱植被指数的分布特征。 通过不同角度图像的NDVI分布情况, 展示监测作物植株不同位置的叶绿素分布情况。 结果显示, RGB-NIR图像匹配时间SIFT(1.865 s)>SURF(1.412 s)>ORB(1.121 s), 匹配准确性上SURF≈SIFT>ORB, 匹配稳定性上SURF>SIFT>ORB, 综合比较选取SURF为最优匹配算法。 采用4灰度级标准板对R, G, B, NIR分量校正模型的R2分别为0.78, 0.76, 0.74, 0.77。 90°和35°视角分别展现了作物叶和茎的NDVI植被指数分布情况, 可为分析和监测作物的营养分布提供技术支持。

The NDVI (Normalized Difference Vegetation Index) calculated based on the spectral reflectance is proved as one of the important parameters to estimate the chlorophyll content of crops, which indicates the growth condition of crop quickly and nondestructively. Thus, the distribution of NDVI of crops can be studied by the binocular stereo vision system with visible RGB (Red, Green, Blue) and near infrared (NIR) images. And the NDVI distribution and dynamics of crops are monitored through the image analysis at different angles. After the spatial distribution maps of crop vegetation index were established based on the matching of RGB and NIR images, the spatial distribution characteristics and influencing factors were discussed by the visualization of NDVI. The RGB and NIR images of 51 maize plants were collected synchronously by the binocular stereo vision system at 90°, 54°, 35° respectively. The RGB-NIR images were pre-processed by Gauss filtering and Laplace operator enhancement. Firstly, three algorithms, namely, SURF (Speeded-Up Robust Features), SIFT (Scale-invariant Feature Transform) and ORB (Oriented Brief), were studied and discussed for RGB-NIR image matching and alignment. Four evaluation indices wereused to determine the optimal matching methodfor RGB-NIR image matching and alignment, including matching time, PSNR (Peak Signal to NoiseRatio), MI (Mutual Information) and SSIM (Structural Similarity Index). Secondly, the crop and background were segmented by using ExG (Extra Green) algorithm and Maximum Interclass Variance algorithm (OTSU). The R (Red), G (Green), B (Blue) and NIR components of the segmented RGB images were extracted. The influence of illumination was discussed and Spectral reflectance was corrected based on the I component of HSI (Hue-Saturation-Intensity) color model. Then, the NDVI of each pixel in the image was calculated, the spatial distribution map of crop vegetation index was drawn, and the distribution characteristics of NDVI under different shooting angles were compared and analyzed. The NDVI distribution was used to display the chlorophyll distribution of crop plants. The RGB-NIR image matching results showed that the matching time with SIFT (1.865 s)>SURF (1.412 s)>ORB (1.121 s), the matching accuracy with SURF≈SIFT>ORB, and the matching stability with SURF≈SIFT>ORB. According to discussion results, the SURF algorithm was selected as the optimal matching algorithm. In order to eliminate the influence of ambient light, the image reflectance was corrected by 4 gray level standard plates on the basis of discussing the I component and gray histogram of HSI model. The R2 of R, G, B and NIR component correction models were 0.78, 0.76, 0.74 and 0.77 respectively. The vegetation index distributions of leaves and stems of crops were presented from 90 and 35 angles, which could provide new technical support for analyzing and monitoring the nutritional status and distribution of crops.