摘要
籽粒的外观特征是区别不同小麦品种的重要标志,对小麦的选育工作具有重要的参考价值.首先采用中值滤波和迭代式阈值法对采集到的4类小麦共468粒样本图像进行处理和分割;然后针对每类小麦,提取了其6个颜色特征、5个形态特征和5个纹理特征等共16个参数;最后通过构建神经网络比较了仅使用颜色和形态特征作为网络输入以及3类特征共同作用时的分类效果.试验结果表明:仅使用颜色、形态两方面的11个特征参数时,小麦样本的识别率为87.6%;当增加5个纹理特征时,样本的识别准确率达到93.13%,可有效识别出4类小麦样本.
The appearance characteristics of grains are the important sign for distinguishing different wheat varieties,and have the important reference value for wheat breeding.Firstly,we acquired the images of 468 samples of 4 kinds of wheat varieties,and processed and partitioned the images by median filtering method and iterative threshold method;then,we extracted 16 parameters including 6 color characteristics,5 morphological characteristics and 5 texture characteristics of each kind of wheat variety;finally,an artificial neural network was constructed to compare the classification effect when using the color and the morphology as the network inputs and during the combined action of the three characteristics.The results showed that the identification rate was 87.6% when only using the 11 characteristic parameters of color and morphology;and identification rate reached 93.13% when the 5 texture characteristic parameters added,so that the 4 kinds of wheat samples can be identified effectively.
出处
《河南工业大学学报（自然科学版）》
CAS
北大核心
2011年第5期74-78,共5页
Journal of Henan University of Technology：Natural Science Edition
关键词
小麦
计算机视觉
特征提取
神经网络
wheat
computer vision
characteristic extraction
artificial neural network
