基于图像分割的玉米粒数高通量估算方法  

Method of High Flux Estimation for Corn Kernels Based on Image Segmentation

作　　者：张颐康 李翠丽[1,2,3] 李源[1,2,3] 孙龙清[1,2,3] 陈英义[1,2,3] 

机构地区：[1]中国农业大学信息与电气工程学院,北京100083 [2]农业部农业信息获取技术重点实验室,北京100083 [3]北京市农业物联网工程技术研究中心,北京100083

出　　处：《湖北农业科学》2016年第12期3185-3188,共4页Hubei Agricultural Sciences

基　　金：国家863计划项目(2013AA10230103)

摘　　要：设计了一种高通量玉米粒计数的方法,首先针对玉米穗分割对边界进行跟踪,根据方向变化幅度寻找"凹点",然后根据边界的方向对"凹点"两两连接进行图像分割;其次提出了改进的支持向量机分类算法的玉米粒识别,除了颜色信息外还将图像中的梯度信息加入分类的自变量中,避免了玉米粒的连通现象;最后通过面积对玉米粒粒数进行估算。结果表明,该方法精确度较高,准确率达到了96.4%;实现了高通量处理,解决了随机摆放的玉米穗粘连分割与高通量条件下图像模糊玉米子粒难以识别的问题。A kind of high flux estimation method was proposed. Firstly, track the boundary and search for the concave by the range ability of direction. Then join every two concave based on the direction in order to segment the corncob. Secondly, use the support vector machine to identify corn kernels based on gradient and color. At last, estimate the number of corn kernels according to area. Experiments show that the estimation accuracy is higher, with 96.4% accuracy. Compared with existing methods, this method simplifies the kernels counting device, realizes the high flux processing, solves the problem of segmenting corncob which is placed randomly and identifying corn kernels which is vague because of the high flux conditions.

关 键 词：图像分割 高通量 最大类间方差法 边界跟踪算法 

分 类 号：S2[农业科学—农业工程] TP339[自动化与计算机技术—计算机系统结构]