小波变换与分水岭算法融合的番茄冠层叶片图像分割

发布时间：2018-03-29 02:02

本文选题：图像分割　切入点：番茄叶片　出处：《农业机械学报》2017年09期

【摘要】：在基于机器视觉的作物营养诊断研究中,通常需要采集叶片样本并在实验室条件下定量测定其营养素含量,但由于叶片间相互重叠,往往使得叶片样本不能清晰地反映在群体番茄冠层图像中。为了解决这一问题,需要利用图像分析技术有效提取作物冠层图像中的叶片,并根据处理结果采集实验室测定样本。本文从复杂背景剔除、梯度图计算、小波变换、标记选取、分水岭分割等环节出发,实现了基于小波变换与分水岭算法融合的番茄冠层多光谱图像叶片分割。首先对比了4种复杂背景剔除算法,发现当增强因子a=1.3时,基于归一化植被指数(Normalized difference vegetation index,NDVI)的阈值分割目标提取准确,适合各种光照条件,时空复杂度低。其次在梯度图计算方面,近红外(Near infrared,NIR)波段图像形态学梯度在保持目标边缘的同时,能消除大量由叶脉、光照等引起的叶片内纹理细节。然后以小波分析为基础进行标记选取,发现当选取db4小波函数、4层小波分解低频系数、阈值为18的H-maxima变换能得到最优的目标标记结果。最后对多光谱番茄冠层图像的小波变换分水岭分割和数学形态学分水岭分割结果进行叠加,发现对复杂背景及不同光照强度下的番茄冠层叶片平均误分率为21%,为基于多光谱图像分析的番茄叶片营养素含量检测提供了一定的技术支持。
[Abstract]:Based on the research of crop nutrition diagnosis of machine vision, usually need to collect leaf samples and determine its nutrient content quantitatively under laboratory conditions, but because the blades overlap each other, often makes the leaf samples could not be clearly reflected in the group of tomato canopy image. In order to solve this problem, the need for effective extraction of leaf crop canopy image by using the technique of image analysis, and according to the results of laboratory samples. The determination of acquisition removed from the complex background, gradient map, wavelet transform, marker selection, and other aspects of a watershed segmentation, the segmentation of tomato canopy based on wavelet transform and watershed algorithm based on multi spectral image. Firstly, leaves compared to the 4 kinds of complex background excluding the algorithm, when the enhancement factor of a=1.3, based on the normalized difference vegetation index (Normalized difference vegetation index, NDVI) threshold segmentation Target extraction accuracy, suitable for all kinds of light conditions, time and space complexity is low. Secondly, in the calculation of the gradient map, near infrared (Near infrared, NIR) band image morphological gradient in the edge of the target at the same time, can be eliminated by a large amount of veins, light induced leaf texture details. Then based on wavelet analysis of mark the selected, found that when the DB4 wavelet function is selected, the 4 layer wavelet decomposition coefficients, the threshold for the 18 H-maxima transform can get the optimal target mark. At the end of the tomato canopy multispectral image wavelet transform and watershed segmentation and mathematical morphology watershed segmentation results are superimposed, tomato canopy on the complicated background and different illumination intensity the average error rate of 21%, provides technical support for the detection of tomato leaf nutrient analysis of multi spectral images based on content.

【作者单位】：兰州城市学院电子与信息工程学院;佛罗里达大学农业与生物工程系;中国农业大学现代精细农业系统集成研究教育部重点实验室;
【基金】：国家自然科学基金项目(31360291、31271619) 国家留学基金委西部地区人才培养特别项目(201408625069) 兰州城市学院博士科研启动基金项目(LZCU-BS2013-07)
【分类号】：S641.2;TP391.41

【相似文献】