基于Fisher鉴别分析和集成学习的烟叶分级方法
Tobacco-leaf Ranking Method Based on Fisher Discriminant Analysis and Ensemble Learning
顾晓东 1陈龙 2王艺斌 3魏新亮 2符再德 4邓斌 2周文辉 2黄振军 2袁志明 2李和平 2刘金云2
作者信息
- 1. 南京理工大学计算机科学与工程学院 南京 210094
- 2. 湖南中烟工业有限责任公司 长沙 430100
- 3. 南京焦耳科技有限责任公司 南京 210032
- 4. 湘西鹤盛原烟发展有限责任公司 吉首 416000
- 折叠
摘要
智能烟叶分级能有效提升成品烟质量.为了进一步提升烟叶分级正确率,提出了联合利用烟叶正面、背面、透视三视图的烟叶分级方法.对于每个视图,提取其纹理、颜色、形状特征作为模型的输入,通过线性判别分析降低特征维度,以支持向量机(Support Vector Machine,SVM)作为基分类器,利用袋装法做集成,得到一种集成学习的烟叶分级模型.通过整合烟叶正面、背面、透视三视图共同提取到特征,带入上述分级模型,进一步提升了分级准确率.实验结果表明,仅对正面视图,提出的集成学习分级模型在五个真实烟叶数据集上的平均准确率为71.93%,优于现有方法.使用三视图分级后,平均准确率提高到了74.8%.
Abstract
Intelligent tobacco-leaf ranking is crucial for the quality of cigarette production.To further improve the accuracy of tobacco-leaf ranking,a method of jointly utilizing three leaf views of front,back and perspective is proposed.For each of the three views,texture,color and shape features of tobacco leaves are extracted as the input of the model,and then the feature dimensionali-ty is reduced by linear discriminant analysis.SVM is used as the base classifier,and multiple SVMs are integrated using bagging method to form a ensemble learning based tobacco ranking model.To integrate the features from the three views,the ensemble mod-el is further improved in ranking accuracy.The experimental results show that for only the front view the average accuracy of the pro-posed method achieves 71.39%on the five real tobacco-leaf datasets,which outperforms several existing methods.After fusing joint features of the three views,the accuracy is up amount to 74.8%.
关键词
烟叶分级/SVM/线性判别分析/集成学习/袋装法Key words
tobacco grading/SVM/linear discriminant analysis/ensemble learning/bagging method引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据