基于分段提取法的日光温室葡萄绿色分数高精度获取

High-precision Acquisition of Grape Green Fraction in Solar Greenhouse Based on Subsection Extraction

为了获取日光温室条件下葡萄植株的高精度绿色分数,利用高通量植物表型监测平台获取葡萄植株全生育期不同生长部位的RGB图像,分别比较迭代法、大津法和直方图双峰法得到其阈值的提取精度,从而筛选确定RGB图像最佳阈值分割方法,并利用差异植被指数(visible-band difference vegetation index,VDVI)、过绿指数(excess green,EXG)和植被颜色提取指数(color index of vegetation,CIVE)共3种植被指数分别结合半径为5像素的圆盘形结构元得到的先闭后开形态学滤波方法,对不同生育期及不同生长部位的绿色分数信息进行提取,并对其提取精度进行分析。结果表明:利用大津法进行阈值分割误差率最小(6.5%),匹配率最高(97.78%),且提取精度高于其他两种方法;通过对葡萄底部冠层图像的绿色分数进行精度比较,发现在萌芽期和抽蔓期,CIVE指数提取精度最高,相对提取误差仅为8.45%和5.27%,而在开花坐果期、浆果膨大期和着色成熟期则采用EXG指数提取效果最佳,提取误差在0.56%~6.39%;在葡萄的不同生长部位,采用CIVE和EXG结合的分段提取法,提取效果均优于单一方法确定的绿色分数,并且在不同部位绿色分数提取精度存在差异,其中冠层中下部、底部精度较高(d>0.944),中上部和顶部稍差,但模型的一次性指数均≥0.940。研究结果可为衡量温室内植株冠层分布提供一种有效途径。

In order to rapidly obtain the high accuracy green fraction of grape plants in solar greenhouse,the high-throughput plant phenotypic monitoring platform was used to obtain RGB images of different locations of plant during the whole growth period.Compare with the Iterative method,Otsu method and Histogram bimodal method respectively were used to obtain the threshold extraction accuracy,so as to screen and determine the optimal threshold segmentation method for RGB images,and then visible-band difference vegetation index(VDVI),excess green(EXG)and color index of vegetation(CIVE)were used,respectively,combined with morphological filtering method of closing first and then opening after obtaining disc-shaped structural elements with a radius of 5 pixels.The green fraction of different growth periods and different zenith angles were extracted,and analyzed its extraction accuracy.The results showed that the error rate of threshold segmentation by Otsu method was the smallest(6.5%)and the matching rate was the highest(97.78%),and the extraction accuracy was higher than those of the other two methods.By comparing the accuracy of the green fraction of grape bottom canopy image,it was found that the CIVE index had the highest extraction accuracy in the germination and the germination period,the relative extraction errors were only 8.45%and 5.27%,while the EXG index was the best in the flowering and fruiting period,the berry expansion period and the coloring stage,extraction errors were 0.56%~6.39%.In different growing parts of grape,the extraction effect of CIVE and EXG was better than that of single method.In different growth parts of grape(zenith angle),the green fraction extraction accuracies were different,in which the accuracies of the lower part and the bottom canopy were higher(d>0.944),the middle and upperpartsand the top canopy were slightly worse,but the one-time index of the model was≥0.940.The results could provide an effective way to measure the distribution of plant canopy in solar greenhouse.