基于多特征融合的花卉种类识别研究

吴笑鑫; 高良; 闫民; 赵方

doi:10.13332/j.1000-1522.20160367

基于多特征融合的花卉种类识别研究

吴笑鑫^1,,
高良¹,
闫民²,
赵方^1,

1.
北京林业大学信息学院
2.
北京林业大学工学院

基金项目:

国家自然科学基金项目 11272061

详细信息

作者简介:
吴笑鑫。主要研究方向：图像处理、模式识别。Email：704124637@qq.com 地址：100083 北京市海淀区清华东路35号北京林业大学信息学院

赵方，副教授。主要研究方向：软件工程、模式识别。Email：929554007@qq.com 地址：同上

中图分类号: S688.9;TP391.4
计量
- 文章访问数: 2638
- HTML全文浏览量: 572
- PDF下载量: 75
出版历程
- 收稿日期: 2016-11-07
- 修回日期: 2016-12-25
- 发布日期: 2017-03-31

Flower species recognition based on fusion of multiple features

1.
School of Information Science and Technology, Beijing Forestry University, Beijing, 100083, P.R. China
2.
School of Technology, Beijing Forestry University, Beijing, 100083, P.R. China

摘要

摘要: 花卉种类识别作为植物自动分类识别的重要分支，有着很高的研究和应用价值。针对当前花卉特征描述存在的局限和花卉识别准确率较低的实际情况，以花卉图像为研究对象，首先对复杂背景图像采用基于显著性检测的Grab Cut分割算法进行预处理，得到单一背景图像；然后在提取花卉图像花冠(所有花瓣)颜色和形状特征的基础上，创新性地提取花蕊区域的颜色和形状所包含的特征信息，并将提取到的18个特征融合成单一特征向量。以支持向量机(SVM)算法为基础构建分类器，通过实验确定核函数与最佳参数；对360幅自建花卉样本库(24个种类，每个种类15幅)进行训练和测试，其中240幅作为训练样本，120幅作为测试样本，并与基于不同特征组合的识别方法进行比较。结果表明：本文提出的基于多特征融合的识别方法具有较高的识别准确率，识别率可以达到92.50%。对通用花卉样本库Oxford 17 flower进行训练与测试，选取其中340幅作为训练样本，170幅作为测试样本，取得了较好的识别效果，验证了本文方法的有效性。
- 花卉图像 /
- 种类识别 /
- 特征提取 /
- 支持向量机
Abstract: As an important branch of plant automatic classification and recognition, flower species recognition is of great value in academic research and practical application. In view of the limitation of description methods for flower features and the problem of low accuracy of flower species recognition, flower images were used as recognition objects, the segmentation algorithm of Grab Cut based on saliency detection was used to preprocess the images with complex background and the images with single background were got. Then the color and shape features of the flower region were extracted. In addition to the color and shape features of the whole flower region, the color and shape features of the pistil/stamen area were also used to represent the flower characteristics more precisely. At last the 18 features were fused into a single feature vector and the SVM (support vector machine) was used for classification and recognition. Experiment based on self-built flower library was conducted for 360 images (15 images each for 24 different species), 240 images were used for the training samples, and the rest of 120 images were used for the test samples. Compared with the recognition methods based on the combination of different features, the method provided in this paper has achieved better recognition effect.The experimental results showed that the recognition accuracy reached 92.50%. Experiments based on Oxford 17 flower library were conducted for 510 images, in which 340 images were used for the training samples, and 170 images were used for the test samples. The experimental results show that the proposed method is effective.
- flower image /
- species recognition /
- feature extraction /
- support vector machine (SVM)

HTML全文

图 1 图像分割过程

Figure 1. Process of image segmentation

下载: 全尺寸图片幻灯片

图 2 12×10的HS空间划分

Figure 2. Separating HS color space into 12×10 color cells

下载: 全尺寸图片幻灯片

图 3 基于多特征融合的花卉种类识别模型

Figure 3. Flower image recognition model based on fusion of multiple features

下载: 全尺寸图片幻灯片

图 4 24类花卉样本

Figure 4. 24 species of flower samples used in experiments

下载: 全尺寸图片幻灯片

图 5 种类7、9、20对比图

Figure 5. Comparison in species 7, species 9 and species 20

下载: 全尺寸图片幻灯片

图 6 不同拍摄角度下的玛格丽特花

Figure 6. Marguerite flowers from different camera angles

下载: 全尺寸图片幻灯片

表 1 不同核函数下的识别率

Table 1 Recognition rate of different kernel functions

训练样本 Training sample	测试样本 Test sample	核函数 Kernel function	识别率 Recognition rate/%
240	120	线性核函数Linear kernel function	84.17
240	120	多项式核函数Polynomial kernel function	82.50
240	120	Sigmoid核函数Sigmoid kernel function	83.33
240	120	径向基核函数Radial basis kernel function	85.83

下载: 导出CSV

表 2 不同特征组合基于不同分类器的识别率

Table 2 Recognition rate of three different classifiers under different combination of features

分类算法Classifying method	CF	SF	PCF	PSF	CF+SF	PCF+PSF	CF+SF+PCF	CF+SF+PSF	CF+SF+PCF+PSF
KNN	44.17	78.33	37.50	34.17	81.67	51.67	82.50	84.17	85.00
BP神经网络BP neutral network	52.50	80.83	43.33	36.67	83.33	57.50	85.83	86.67	87.50
SVM	58.33	85.00	46.67	40.83	86.67	63.33	88.33	89.17	92.50
注：CF代表花冠颜色特征，SF代表花冠形状特征，PCF代表花蕊区域颜色特征，PSF代表花蕊区域形状特征。Notes: CF represents corolla color feature, SF represents corolla shape feature, PCF represents color feature in pistil area, PSF represents shape feature in pistil area.

下载: 导出CSV

表 3 不同方法对每类花卉的识别率

Table 3 Recognition rate for each class of flower based on different methods

类别 Flower species	识别率1 Recognition rate 1/%	识别率2 Recognition rate 2/%
1	100	100
2	100	100
3	80	80
4	80	80
5	80	100
6	100	100
7	60	80
8	100	100
9	60	80
10	100	100
11	80	80
12	80	100
13	100	100
14	100	100
15	80	80
16	80	100
17	100	100
18	100	100
19	80	80
20	40	80
21	100	100
22	100	100
23	100	100
24	80	80
注：识别率1，基于花冠颜色特征与花冠形状特征的识别率；识别率2，基于花冠颜色特征、花冠形状特征和花蕊区域特征的识别率。Notes: recognition rate 1, recognition rate based on color and shape features of corolla region; recognition rate 2, recognition rate based on color and shape features of corolla region and pistil area.

下载: 导出CSV

表 4 基于不同方法的花卉识别率

Table 4 Recognition rate based on different methods

%
采用方法Method	识别率Recognition rate
本文方法Method proposed	80.0
文献[5] Literature [5]	72.8
文献[6] Literature [6]	76.3
文献[7] Literature [7]	79.4
文献[8] Literature [8]	80.7

下载: 导出CSV

参考文献(16)

[1]	柯逍, 陈小芬, 李绍滋.基于多特征融合的花卉图像检索[J].计算机科学, 2010, 37(11):282-286. doi: 10.3969/j.issn.1002-137X.2010.11.068 KE X, CHEN X F, LI S Z. Flower image retrieval based on multi-features fusion[J]. Computer Science, 2010, 37(11): 282-286. doi: 10.3969/j.issn.1002-137X.2010.11.068
[2]	周伟, 武港山.基于显著图的花卉图像分类算法研究[J].计算机技术与发展, 2011, 21(11):15-18. doi: 10.3969/j.issn.1673-629X.2011.11.005 ZHOU W, WU G S. Research on saliency map based flower image classification algorithm[J]. Computer Technology and Development, 2011, 21(11):15-18. doi: 10.3969/j.issn.1673-629X.2011.11.005
[3]	张娟, 黄心渊.基于图像分析的梅花品种识别研究[J].北京林业大学学报, 2012, 34(1):96-104. http://j.bjfu.edu.cn/article/id/9711 ZHANG J, HUANG X Y. Species identification of Prunus mume based on image analysis[J]. Journal of Beijing Forestry University, 2012, 34(1):96-104. http://j.bjfu.edu.cn/article/id/9711
[4]	白帆, 郑慧峰, 沈平平, 等.基于花朵特征编码归类的植物种类识别方法[J].浙江大学学报(工学版), 2015, 49(10):1902-1908. http://d.old.wanfangdata.com.cn/Periodical/zjdxxb-gx201510011 BAI F, ZHENG H F, SHEN P P, et al. Plant species identification method based on flower feature coding classification[J]. Journal of Zhejiang University(Engineering Science), 2015, 49(10):1902-1908. http://d.old.wanfangdata.com.cn/Periodical/zjdxxb-gx201510011
[5]	NILSBACK M E, ZISSERMAN A. Automated flower classification over a large number of classes[C]//Proceedings of Indian Conference on Computer Vision, Graphics & Image Processing. Bhubaneswar: IEEE, 2008: 722-729.
[6]	NILSBACK M E, ZISSERMAN A. Delving deeper into the whorl of flower segmentation[J]. Image & Vision Computing, 2010, 28(6):1049-1062. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=8b2bd882ca3a14c96af35700d88163ca
[7]	CHAI Y, LEMPITSKY V, ZISSERMAN A. BiCoS: a Bi-level co-segmentation method for image classification[C]//Proceedings of IEEE International Conference on Computer Vision. Barcelona: IEEE, 2011: 2579-2586.
[8]	ANGELOVA A, ZHU S. Efficient object detection and segmentation for fine-grained recognition[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. Washington: IEEE Computer Society, 2013: 811-818.
[9]	YANIKOGLU B, YANIKOGLU B, APTOULA E, et al. Sabanci-Okan system at ImageClef 2012: combining features and classifiers for plant identification[C]. Rome: CLEF, 2012.
[10]	王丽君, 淮永建, 彭月橙.基于叶片图像多特征融合的观叶植物种类识别[J].北京林业大学学报, 2015, 37(1):55-61. doi: 10.13332/j.cnki.jbfu.2015.01.006 WANG L J, HUAI Y J, PENG Y C. Method of identification of foliage from plants based on extraction of multiple features of leaf images[J]. Journal of Beijing Forestry University, 2015, 37(1):55-61. doi: 10.13332/j.cnki.jbfu.2015.01.006
[11]	ACHANTA R, SHAJI A, SMITH K, et al. SLIC superpixels compared to state-of-the-art superpixel methods.[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2012, 34(11):2274-2282. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=218e2dd693ea07f7be3f296ed4e6aaba
[12]	YANG C, ZHANG L, LU H, et al. Saliency detection via graph-based manifold ranking[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. Washington: IEEE Computer Society, 2013: 3166-3173.
[13]	ROTHER C, KOLMOGOROV V, BLAKE A. GrabCut: interactive foreground extraction using iterated graph cuts[J]. Acm Transactions on Graphics, 2004, 23(3):309-314. doi: 10.1145/1015706.1015720
[14]	裴勇.基于数字图像的花卉种类识别技术研究[D].北京: 北京林业大学, 2011. http://cdmd.cnki.com.cn/Article/CDMD-10022-1011134451.htm PEI Y. Study on the flower identification technology with digital images[D].Beijing: Beijing Forestry University, 2011. http://cdmd.cnki.com.cn/Article/CDMD-10022-1011134451.htm
[15]	GONZALEZ R C, WINTZ P. Digital image processing[J]. Prentice Hall International, 2002, 28(4):484-486. http://d.old.wanfangdata.com.cn/Periodical/bjykdxxx201601025
[16]	张学工.关于统计学习理论与支持向量机[J].自动化学报, 2000, 26(1):32-42. doi: 10.3969/j.issn.1003-8930.2000.01.008 ZHANG X G. Introduction to statistical learning theory and support vector machines[J]. Acta Automatica Sinica, 2000, 26(1):32-42. doi: 10.3969/j.issn.1003-8930.2000.01.008

施引文献(7)

期刊类型引用(6)

1.	赵钰婷，陈冬瑶，杨柳，李晶楠，宁广亮，姜静. 白桦四倍体×紫雨桦二倍体杂交种子活力及杂种子代生长特性分析. 温带林业研究. 2025(01): 1-8 . 百度学术
2.	刘笑，杜琬莹，张云秀，唐成名，李华伟，夏海勇，樊守金，孔令安. NO_3~–缓解小麦根部NH_4~+毒性机理. 植物学报. 2024(03): 397-413 . 百度学术
3.	魏尚霖，李晨蕾，廖柏勇，程俊森，代文魁，姜维，王溢，李永泉. 高州油茶与小果油茶种仁蛋白质组差异比较. 经济林研究. 2024(03): 10-24 . 百度学术
4.	程俊森，王溢，李永泉，魏尚霖，李超楠，姜维，黄润生. 基于Label-free技术的高州油茶铝胁迫蛋白质组学研究. 中南林业科技大学学报. 2023(08): 169-181 . 百度学术
5.	韦庆钰，黄海龙，吴纯泽，苏嘉熙，卫星. 3种倍性青杨扦插苗对覆膜滴肥的生长响应. 南京林业大学学报(自然科学版). 2021(05): 93-101 . 百度学术
6.	王溢. 基于label-free技术的青杨3个叶位叶片比较蛋白质组学分析. 华中农业大学学报. 2019(04): 8-19 . 百度学术