摘要
针对单依靠颜色或形状将采摘期玫瑰花从图像中分割出来难度较大的问题,研究一种基于神经网络的食用玫瑰花图像识别算法。将处于采摘期的玫瑰花正面图像作为识别对象,先提取HSI色彩空间下的S分量,用最大类间方差法(Otsu)进行分割;再提取目标图像灰度共生矩阵下的纹理特征,选取区分度高的纹理特征,结合BP神经网络,建立识别模型。试验结果表明:该方法正确识别率>85%,识别率主要受试验样本开放标准选取的影响,而受光照影响不敏感,是一种较好的识别方法。
It is quite difficult to distinguish the appropriate rose by its color or shape.In order to solve this problem,this paper is aimed to explore an image segmentation and recognition algorithm of edible rose.The roses just before picking period were taken as the recognition object.Firstly,shifted this image to HIS color space;extracted S weight and segmented the image by using Otsu method.Then extracted the textural features of high distinction degree and recognized the picking rose by integrating BP neural network.The result indicated that the correct recognition rate of this method was higher than 85%and the recognition rate was mainly affected by the samples' openness instead of their sensitive to the illumination.
出处
《中国农业大学学报》
CAS
CSCD
北大核心
2014年第4期180-186,共7页
Journal of China Agricultural University
基金
国家自然科学基金项目(31060118)
云南省应用基础研究项目(2009ZC041M)
昆明理工大学人才培养项目(2010-07)
关键词
食用玫瑰花
图像分割
纹理特征
神经网络
edible rose
image segmentation
textural features
neural network
