光谱指数测量中窄带图像的光强自适应分割方法

Light intensity self-adaption segmentation method of narrow-band image in measuring spectral index

[目的]本文对作物冠层区域的光强自适应性分割算法进行了研究,旨在消除光谱指数测量过程中土壤背景的干扰,克服现有光谱检测方法对结构光的依赖性。[方法]采用加装窄带滤光片的2个相机组建测量系统,并对相机的几何安装参数进行标定;将窄带图像灰度直方图的波谷灰度值与最大灰度值的比值作为归一化阈值,实现窄带图像分割的光环境自适应性。在实现近红外窄带图像中土壤和冠层分割的基础上,通过几何变换推算出可见光波段窄带图像的冠层区域,从而实现可见光窄带图像的分割。[结果]在80~140 cm高度内,每隔10 cm设置1个测量高度,每个高度采集5组平均株高为25 cm的绿萝冠层窄带图像(770和660 nm),安装660 nm滤光片的相机在完成图像采集后,将滤光片换成770 nm重新拍摄5张照片,作为660 nm图像分割效果的参考图像。结果表明:660 nm图像的分割区域与参考图像的分割区域平均重合度大于99%。同时计算了光照度为10000~26000 lx时拍摄的50幅绿萝770 nm窄带图像的非归一化阈值和归一化阈值,其与光照度的相关系数分别为0.5866和0.0916。[结论]本文提出的基于归一化阈值的分割方法对光环境的变化具有很强的适应性,综合归一化阈值和几何变换可以实现窄带图像的分割,为构建消除土壤干扰的归一化植被指数(normalized difference vegetation index,NDVI)等光谱指数提供有效的基础数据。

[Objectives]In order to eliminate the interference of soil background in the process of spectral indices measurement and overcome the dependence of existing spectral detection methods on structural light,the self-adaption segmentation algorithm of crop canopy region was studied in this paper.[Methods]In this paper,two cameras with narrow-band filters were used to build the measurement system,and the geometric installation parameters of the two cameras were calibrated.The ratio of the valley gray value of the narrow-band image gray histogram to the maximum gray value was used as a normalization threshold to achieve optical environment adaptability of narrow-band image segmentation.On the basis of realizing the segmentation of soil and canopy in the near-infrared narrow-band image,the canopy regions of the narrow-band images in the visible light band were derived by geometric transformation,and the narrow-band image of visible light could be segmented.[Results]In the height range of 80-140 cm,a measurement height was set every 10 cm,and each group collected 5 sets of narrow-band images(770 and 660 nm)of green radish with an average height of 25 cm.After the image acquisition was completed,the camera with the 660 nm filter was replaced with 770 nm to retake 5 photos as the reference images for the 660 nm images’segmentation effect.The results showed that each average overlap of the segmentation region of the 660 nm image and the segmentation region of the reference image was greater than 99%.At the same time,the non-normalized thresholds and normalized thresholds of the 770 nm narrow-band images of 50 green radish photographed in the range of 10000-26000 lx were calculated,and their correlation coefficients with light intensity were 0.5866 and 0.0916,respectively.[Conclusions]The segmentation method based on normalized threshold proposed in this paper has strong adaptability to the changes of light environment.Combined with normalized threshold and geometric transformation,it can realize the segmentation of narrow-band images and provide effective basic data for constructing spectral indices which eliminate soil interference such as normalized difference vegetation index(NDVI).