摘要
针对水稻种子相似度高、识别困难等问题,提出一种线性判别分析(Linear discriminant analysis,LDA)和贝叶斯分类(Bayes)相结合的分类识别方法,以提高水稻种子分类识别速度和识别准确率。通过对4类水稻种子(楚粳7号、马坝油粘、玉杨糯、玉针香)的图像进行裁剪和分割等预处理操作,提取出水稻种子图像的颜色特征、几何特征和纹理特征。利用线性判别分析、主成分分析、因子分析和局部线性嵌入对特征数据进行分析降维,并分别选择Bayes、K-邻近、支持向量机、多层感知机分类器对原始特征数据和降维数据进行分类识别研究。为提高模型泛化能力,通过图像增强技术对稻种原始数据集进行样本扩充,利用图像增强技术模拟多种环境对水稻种子图片数据集进行增强处理,结果显示,基于数据增强后的LDA_Bayes模型运行时间为0.019 s,识别准确率为99.4%。与其他模型比较,该模型具有更强的鲁棒性和适用性,能高效地分类识别不同环境下的水稻种子,可为水稻种子分类识别提供一种新方法。
Aiming at the problems of high similarity and difficult identification of rice seeds,a classification and identification method which combined linear discriminant analysis(LDA)and Bayesian classification(Bayes)was proposed to improve the identification speed and accuracy of rice seed classification.By performing pre-processing operations such as cropping and segmentation on the images of four types of rice seeds(Chujing No.7,Maba Younian,Yuyangnuo and Yuzhenxiang),the color characteristics,geometric and texture features of rice seed images were extracted.Linear discriminant analysis,principal component analysis,factor analysis and locally linear embedding were used to analyze and reduce the dimensionality of the feature data,and Bayes,K-nearest neighbors,support vector machine,multilayer perceptron classifiers were selected respectively to conduct classification and identification research on the original feature data and dimensionality reduced data.To improve the generalization ability of the model,image enhancement technology was used to extend the original data set of rice seed image samples and to simulate multiple environments to enhance the image data set of rice seeds.The results showed that the running time of LDA_Bayes model based on data enhancement was 0.019 s and the recognition accuracy rate was 99.4%.Compared with other models,LDA_Bayes model shows stronger robustness and applicability,it can classify and recognize rice seeds in different environments efficiently and provide a new method for rice seed classification and identification.
作者
黄琼
杨红云
万颖
HUANG Qiong;YANG Hong-yun;WAN Ying(College of Computer and Information Engineering, Jiangxi Agricultural University, Nanchang 330045, China;College of Software, Jiangxi Agricultural University, Nanchang 330045, China)
出处
《江苏农业学报》
CSCD
北大核心
2021年第1期8-15,共8页
Jiangsu Journal of Agricultural Sciences
基金
国家自然科学基金项目(61562039)。
关键词
水稻种子
线性判别
BAYES
分类识别
rice seeds
linear discrimination
Bayes
classification and recognition
