基于改进SLIC的光照干扰下茶树冠层图像分割  

Segmentation of tea plant canopy image under light interference based on improved SLIC

作　　者：刘连忠[1] 李孟杰 宁井铭 LIU Lian-zhong;LI Meng-jie;NING Jing-ming(School of Information and Computer, Anhui Agricultural University, Hefei 230036, China;State Key Laboratory of Tea Plant Biology and Resource Utilization, Hefei 230036, China)

机构地区：[1]安徽农业大学信息与计算机学院,安徽合肥230036 [2]茶树生物学与资源利用国家重点实验室,安徽合肥230036

出　　处：《江苏农业学报》2020年第4期1022-1027,共6页Jiangsu Journal of Agricultural Sciences

基　　金：安徽省高校自然科学研究重点项目(KJ2017A151);茶树生物学与资源利用国家重点实验室开放基金项目(SKLTOF20160202);安徽高校自然科学研究重大项目(KJ2019ZD20);国家重点研发计划项目(2016YFD0200900)。

摘　　要：针对自然环境下获取农作物图像时极易受到光照干扰的问题,提出一种改进的简单线性迭代聚类(SLIC)方法,以[L^*,R,G-S,x,y]作为聚类向量对茶树冠层图像进行超像素分割,提取超像素块的R、G、B、H、S、V、L^*、a^*、b^*、熵、能量、对比度、逆差矩等13个图像特征参数;将超像素块分为正常区域、反光区域、背景3类,分别选择线性、多项式和RBF核函数的SVM进行分类,得到仅包含正常区域的茶树冠层图像,进而提取正常区域的图像特征参数。试验结果表明,在光照变化情况下,改进的SLIC与RBF-SVM结合得到的图像特征最为稳定。In order to solve the problem that crop images in natural environment were easily affected by the change of light,the improved simple linear iterative clustering(SLIC)method was proposed.Using[L^*,R,G-S,x,y]as clustering vector,the superpixel segmentation of the tea plant canopy image was carried out.Then 13 image feature parameters such as R,G,B,H,S,V,L^*,a^*,b^*,entropy,energy,contrast and inverse difference moment of the super-pixel block were extracted.The super-pixel blocks were divided into three categories:normal area,reflective area and background,and classified by support vector machine(SVM)with linear,polynomial and RBF kernel functions separately.Finally,the canopy image of tea plant containing only the normal area was obtained,and the image feature parameters of the normal area were extracted.The results showed that the image feature parameters obtained by the improved SLIC and RBF-SVM were the most stable under the condition of light interference.

关 键 词：农作物图像 图像分割 超像素 图像分类 特征参数 

分 类 号：S126[农业科学—农业基础科学]