摘要
为了实现对空气中的致敏花粉信息进行自动化统计,针对上海地区典型气传致敏花粉的光学显微镜图像,提出了基于形状和纹理特征的识别方法。对图像中分割得到的花粉区域,使用全局形状描述和傅里叶描述子提取形状信息,灰度共生矩阵提取纹理特征,并且构建k近邻分类器进行识别。选用桑科56例、禾本科25例和松科60例共141例实验样本,分别可以实现91%、88%和98%分类准确率。实验结果表明,该方法可以初步实现对花粉显微图像的分割和识别,为花粉的自动识别系统打下基础。
In order to realize the automatic statistic of allergic pollen grains in air,by identifying airborne pollen grains in optical micro-scope images typically for Shanghai area,a method based on shape and texture is proposed.After pollen region is segmented,global shape description and Fourier descriptor is used to extract shape feature,gray level co-occurrence matrix is used to extract texture feature,finally the pollen grains are classified by a k-nearest neighbor classifier.In experiment with 56 cases of Moraceae,25 cases of Poaceae and 60 cases of Pinaceae,an accuracy of 91%,88% and 98% for each can be reached.Finally it is indicated that initial recognize is realized,which lay a basis for the auto-recognize system of allergic pollen.
出处
《计算机工程与设计》
CSCD
北大核心
2011年第4期1379-1382,共4页
Computer Engineering and Design
关键词
气传致敏花粉
显微图像
全局形状描述
傅里叶描述子
灰度共生矩阵
K近邻
airborne pollen
microscopy image
global shape description
Fourier descriptor
gray level co-occurrence matrix
k-nearest neighbor classifier
