Influence of Image Features and Sample Sizes on Rice Pest Identification

doi:10.16819/j.1001-7216.2018.7116

Abstract

Abstract:

【Objective】In the traditional pattern recognition methods, image features and the sizes of training samples have a great influence on the identification results of target objects from a large number of distraction objects. Our objective is to study the influence of different image features and sample sizes on identification of rice light-trapped pests. 【Methods】 Rice light-trapped insects were divided into two broad categories：big insects and small insects. The global and local image features of all insects were extracted and different sizes of training samples were set to train support vector machine classifiers. 【Result】The support vector machine classifier based on the combination of global features and HOG features could obtain the identification rate of 91.4% and false detection rate of 8.6% when the non-target sample size was fourfold as many as target samples in big rice pests. The support vector machine classifier based on global features could obtain the identification rate of 94.9% and false detection rate of 4.9% when the non-target sample size was two times as many as target samples in small rice pests. 【Conclusion】In the small sample sets, appropriate image features and reasonable training sample proportion help achieve good identification results when some targets need to be identified from a large number of non-target objects.

Key words: rice pest, pattern recognition, global feature, local feature, training sample, support vector machine

摘要：

【目的】在传统的模式识别分类中,从大量的干扰物体中识别出目标物体,图像特征参数的选择和不同训练样本数量的比例对目标物体的识别结果有着较大的影响。研究的目的在于明晰不同的图像特征和样本量对水稻灯诱害虫识别结果的影响。【方法】根据5种目标害虫体型大小,将水稻灯诱昆虫分成大型昆虫和小型昆虫。研究水稻昆虫图像的全局特征、局部特征和不同特征融合对水稻目标害虫识别结果的影响;研究基于小样本条件,选择不同训练样本比例对水稻目标害虫识别结果的影响。【结果】当非目标昆虫样本量约为目标害虫样本量的4倍时,基于全局特征和HOG特征融合训练得到的支持向量机分类器识别水稻3种大型害虫,可获得91.4%的识别率和8.6%的误检率;当非目标昆虫样本量约为目标害虫样本量的2倍左右时,基于全局特征的支持向量机分类器识别水稻2种小型害虫,可获得94.9%的识别率和4.9%的误检率。【结论】针对小样本数据,从大量非目标中识别出目标物体,选择合适的特征和设置合理的训练样本比例可获得较好的目标识别结果。

关键词: 水稻害虫, 模式识别, 全局特征, 局部特征, 训练样本, 支持向量机

CLC Number:

Pengpeng MA, Aiming ZHOU, Qing YAO, Baojun YANG, Jian TANG, Xiuqiang PAN. Influence of Image Features and Sample Sizes on Rice Pest Identification[J]. Chinese Journal OF Rice Science, 2018, 32(4): 405-414.

马鹏鹏, 周爱明, 姚青, 杨保军, 唐健, 潘修强. 图像特征和样本量对水稻害虫识别结果的影响[J]. 中国水稻科学, 2018, 32(4): 405-414.

Figures/Tables 9

References 41

[1]	刘芳, 沈佐锐, 张建伟, 杨红珍. 基于颜色特征的昆虫自动鉴定方法. 应用昆虫学报, 2008, 45(1): 150-153.
	Liu F, Shen Z R, Zhang J W, Yang H Z.Automatic insect identification based on color characters.Chin Bull Entomol, 2008, 45(1): 150-153. (in Chinese with English abstract)
[2]	于新文, 沈佐锐, 高灵旺, 李志红. 昆虫图像几何形状特征的提取技术研究. 中国农业大学学报, 2003, 8(3): 47-50.
	Yu X W, Shen Z R, Gao L W, Li Z H.Feature measuring and extraction for digital image of insects.J China Agric Univ, 2003, 8(3): 47-50. (in Chinese with English abstract)
[3]	张红涛, 胡玉霞, 赵明茜, 邱道尹, 张孝远, 张恒源. 田间害虫图像识别中的特征提取与分类器设计研究. 河南农业科学, 2008, 37(9): 73-75.
	Zhang H T, Hu Y X, Zhao M Q, Qiu D Y, Zhang X Y, Zhang H Y.Feature Extraction and classification in the image recognition for agricultural pests.J Henan Agric Sci, 2008, 37(9): 73-75. (in Chinese)
[4]	赵三琴, 丁为民, 刘德营. 基于傅里叶描述子的稻飞虱形状识别. 农业机械学报, 2009, 40(8): 181-184.
	Zhao S Q, Ding W M, Liu D Y.Rice hopper shape recognition based on Furier descriptors.Trans Chin Soc Agric Mach, 2009, 40(8): 181-184. (in Chinese with English abstract)
[5]	Zhao J, Cheng X P.Field pest identification by an improved Gabor texture segmentation scheme.New Zeal J Agric Res, 2007, 50(5): 719-723.
[6]	Larios N, Deng H, Zhang W, Sarpola M, Yuen J, Paasch R, Moldenke A, Lytle D.A, Correa S.R, Mortensen E, Shapiro L G, Dietterich T G. Automated insect identification through concatenated histograms of local appearance features: Feature vector generation and region detection for deformable objects.Mach Vision & Appl, 2008, 19(2): 105-123.
[7]	Wen C, Guyer D E, Li W.Automated insect classification with combined Global and local features for orchard management//An ASABE Meeting Presentation, June 21-24, 2009. Reno, Nevada: 2009.
[8]	李凡. 基于数字图像的蝴蝶种类自动识别研究. 北京:北京林业大学, 2015.
	Li F.The Research on Automatic Identification of Butterfly Species Based on the Digital Image. Beijing: Beijing Forestry University, 2015. (in Chinese with English abstract)
[9]	胡永强, 宋良图, 张洁, 谢成军, 李瑞. 基于稀疏表示的多特征融合害虫图像识别. 模式识别与人工智能, 2014, 27(11): 985-992.
	Hu Y Q, Song L T, Zhang J, Xie C J, Li R.Pest Image recognition of multi-feature fusion based on sparse representation. Pat Rec & Art Int, 2014, 27(11): 985-992. (in Chinese with English abstract)
[10]	Liu Z, Gao J, Yang G, Zhang H, He Y.Localization and classification of paddy field pests using a saliency map and deep convolutional neural network.Sci Rep, 2016, 6: 20410.
[11]	杨国国, 鲍一丹, 刘子毅. 基于图像显著性分析与卷积神经网络的茶园害虫定位与识别. 农业工程学报, 2017, 33(6): 156-162.
	Yang G G, Bao Y D, Liu Z Y.Location and recognition of pests in tea plantation based on image saliency analysis and convolutional neural network.Trans Chin Soc Agric Eng, 2017, 33(6): 156-162. (in Chinese with English abstract)
[12]	Wen C, Wu D, Hu H, Pan W.Pose estimation-dependent identification method for field moth images using deep learning architecture.Bios Eng, 2015, 136: 117-128.
[13]	Lytle D A, Martínezmuñoz G, Zhang W, Lorios N, Shaprio L.Automated processing and identification of benthic invertebrate samples. J North Am Benthol Soc, 2010, 29(3): 867-874.
[14]	冼鼎翔, 姚青, 杨保军, 罗举, 谭畅, 张超, 徐一成. 基于图像的水稻灯诱害虫自动识别技术的研究. 中国水稻科学, 2015, 29(3): 299-304.
	Xian D X, Yao Q, Yang B J, Luo J, Tan C, Zhang C, Xu Y C.Automatic identification of rice light-trapped pest based on image.Chin J Rice Sci, 2015, 29(3): 299-304. (in Chinese with English abstract)
[15]	刘鹏宇. 基于内容的图像特征提取算法的研究. 长春: 吉林大学, 2004.
	Liu P Y.Study on the algorithms for content-based image feature extraction. Changchun: Jilin University, 2004. (in Chinese with English abstract)
[16]	梁勇. 智能虫情测报灯在水稻害虫测报上的应用效果. 广西植保, 2009, 22(1): 21-23.
	Liang Y.Application effect of intelligent insect report lamp on rice pest monitoring.Guangxi Plant Protect, 2009, 22(1): 21-23. (in Chinese)
[17]	张新明, 张爱丽, 郑延斌, 孙印杰, 李双. 改进的最大熵阈值分割及其快速实现. 计算机科学, 2011, 38(8): 278-283.
	Zhang X M, Zhang A L, Zheng Y B, Sun Y J, Li S.Improved two-dimensional maximum entropy image thresholding and its fast recursive realization.Comput Sci, 2011, 38(8): 278-283. (in Chinese with English abstract)
[18]	苏煜, 山世光, 陈熙霖, 高文. 基于全局和局部特征集成的人脸识别. 软件学报, 2010, 21(8): 1849-1862.
	Su Y, Shan S G, Chen X L, Gao W.Integration of global and local feature for face recognition. J Softw, 2010, 21(8): 1849-1862. (in Chinese with English abstract)
[19]	彭晶. 基于全局特征点提取的指纹识别方法研究. 长春: 吉林大学, 2005.
	Peng J.The study of fingerprint identification based on the deltas and cores. Changchun: Jilin University, 2005. (in Chinese with English abstract)
[20]	Liu B, Wu H, Su W, Sun J.Sector-ring HOG for rotation-invariant human detection.Signal Proc Image Comm, 2017, 54(C): 1-10.
[21]	Zhu J X, Su G D, Li Y C.Facial expression recognition based on gabor feature and adaboost.Mod Sci Instrum, 2011, 17(8): 993-998.
[22]	Lei Y M, Zhao X M, Guo W D.Cirrhosis recognition of liver ultrasound images based on SVM and uniform LBP feature//2015 IEEE Advanced Information Technology, Electronic and Automation Control Conference (IAEAC), 19-20 December, 2015. Chongqing, China: IEEE, 2015: 382-387.
[23]	Olgun M, Onarcan A O, Sezer O, Koyuncu O.Wheat grain classification by using dense SIFT features with SVM classifier.Comput & Elec Agric, 2016, 122(C): 185-190.
[24]	Dlin S, Jiang Y T, Lin J J.Object Tracking Using a Particle Filter with SURF Feature.J Elec Sci & Technol, 2014, 8815(3): 339-344.
[25]	王娟, 孔兵, 贾巧丽. 基于颜色特征的图像检索技术. 计算机系统应用, 2011, 20(7): 160-164.
	Wang J, Kong B, Jia Q L.Color-Based Image Retrieval.Comut Sys Appl, 2011, 20(7): 160-164. (in Chinese with English abstract)
[26]	任国贞, 江涛. 基于灰度共生矩阵的纹理提取方法研究. 计算机应用与软件, 2014(11): 190-192.
	Ren G Z, Jiang T.Study on glcm-based texture extraction methods.Comput Appl Softw, 2014(11): 190-192. (in Chinese with English abstract)
[27]	郑淑丹, 郑江华, 石明辉, 郭宝林, 森巴提, 孙志群, 贾晓光, 李晓瑾. 基于分形和灰度共生矩阵纹理特征的种植型药用植物遥感分类. 遥感学报, 2014, 18(4): 868-886.
	Zheng S D, Zheng J H, Shi M H, Guo B L, Sen B T, Sun Z Q, Jia X G, Li X J.Classification of cultivated Chinese medicinal plants based on fractal theory and gray level co-occurrence matrix textures.J Remot Sens, 2014, 18(4): 868-886. (in Chinese with English abstract)
[28]	杨舒, 王玉德. 基于Contourlet变换和Hu不变矩的图像检索算法. 红外与激光工程, 2014, 43(1): 306-310.
	Yang S, Wang Y D.Image retrieval algorithm based on contourlet transform and Hu invariant moment.Infrar Laser Eng, 2014, 43(1): 306-310. (in Chinese with English abstract)
[29]	张华. 基于形态特征提取的图像匹配搜索技术研究. 物联网技术, 2013(11): 16-18.
	Zhang H.Research on image matching search technology based on morphology characterization extraction.Internet Things Technol, 2013(11): 16-18. (in Chinese)
[30]	Arivazhagan S, Ganesan L, Priyal S P.Texture classification using Gabor wavelets based rotation invariant features.Patt Recog Lett, 2006, 27(16): 1976-1982.
[31]	王昊京, 王建立, 王鸣浩, 阴玉梅. 采用双线性插值收缩的图像修复方法. 光学精密工程, 2010, 18(5): 1234-1241.
	Wang H J, Wang J L, Wang M H, Yin Y M.Efficient image inpainting based on bilinear interpolation downscaling.Optics Prec Eng, 2010, 18(5): 1234-1241. (in Chinese with English abstract)
[32]	Ojala T, Pietikainen M, Maenpaa T.Multiresolution gray-scale and rotation invariant texture classification with local binary patterns// IEEE Transactions on Pattern Analysis and Machine Intelligence.Springer Berlin Heidelberg, 2000: 404-420.
[33]	曾凡涛. 基于改进LBP特征的图像理解. 长春: 吉林大学, 2014.
	Zeng F T.Image Understanding Based on Improved LBP Methodology. Changchun: Jilin University, 2014. (in Chinese with English abstract)
[34]	陈振华, 余永权, 张瑞. 模糊模式识别的几种基本模型研究. 计算机技术与发展, 2010, 20(9): 32-35.
	Chen Z H, Yu Y Q, Zhang R.Research on several models of fuzzy pattern recognition problems.Comput Technol Dev, 2010, 20(9): 32-35. (in Chinese with English abstract)
[35]	王双成, 杜瑞杰, 刘颖. 连续属性完全贝叶斯分类器的学习与优化. 计算机学报, 2012, 35(10): 2129-2138.
	Wang S C, Du R J, Liu Y.The learning and optimization of full bayes classifiers with continuous attributes. Chin J Comput, 2012, 35(10): 2129-2138. (in Chinese with English abstract)
[36]	吴晓军, 邹广华. 基于边缘几何特征的高性能模板匹配算法. 仪器仪表学报, 2013, 34(7): 23-30.
	Wu X J, Zou G H.High performance template matching algorithm based on edge geometry features.Chin J Sci Instrum, 2013, 34(7): 23-30. (in Chinese with English abstract)
[37]	罗昕. 基于Fisher鉴别分析对支持向量机算法进行优化的研究. 昆明: 云南大学, 2016.
	Luo X.Research on the optimization of support vector machine algorithm based on Fisher discriminant analysis. Kunming: Yunnan University, 2016. (in Chinese with English abstract)
[38]	任俊, 李志能. 支持向量机在字符分类识别中的应用. 浙江大学学报: 工学版, 2005, 39(8):1136-1141.
	Ren J, Li Z N.Application of support vector machines in classification and recognition of characters.J Zhejiang Univ: Eng Sci, 2005, 39(8): 1136-1141. (in Chinese with English abstract)
[39]	崔萌, 张春雷,. LIBSVM, LIBLINEAR, SVMmuticlass比较研究. 电子技术, 2015(6): 1-5.
	Cui M, Zhang C L.The comparison study of LIBSVM, LIBLINEAR and SVMmulticlass.Elec Technol, 2015 (6):1-5. (in Chinese)
[40]	邓江洪, 赵领. 多特征筛选与支持向量机相融合的图像分类模型. 吉林大学学报:理学版, 2016, 54(4): 862-866.
	Deng J H, Zhao L.Image Classification model with multiple feature selection and support vector machine.J Jilin Univ: Sci Ed, 2016, 54(4): 862-866. (in Chinese with English abstract)
[41]	沙睿. 基于多特征融合的图像相似性研究. 南京: 南京大学, 2011.
	Sha R.A Study of image Similarity based on multi-features. Nanjing: Nanjing University, 2011. (in Chinese with English abstract)

样本分类 Material category		昆虫姿态 Insect posture	用于训练的样本最大数 Max sample size of training images	测试样本数 Sample size of testing images
大型昆虫 Big insects	大螟 Sesamia inferens	背面Back image	330	140
	大螟 Sesamia inferens	腹面Abdomen image	428	140
	稻纵卷叶螟 Cnaphalocrocis medinalis	背面Back image	355	140
	稻纵卷叶螟 Cnaphalocrocis medinalis	腹面Abdomen image	364	140
	二化螟 Chilo suppressalis	背面Back image	355	140
	二化螟 Chilo suppressalis	腹面Abdomen image	364	140
	大型非目标昆虫Big non-target insects		5800	140
小型昆虫 Small insects	白背飞虱Sogatella furcifera	侧面Lateral image	3200	215
	褐飞虱Nilaparvata lugens	侧面Lateral image	800	215
	小型非目标昆虫Small non-target insects		6000	215

样本分类 Material category		昆虫姿态 Insect posture	用于训练的样本最大数 Max sample size of training images	测试样本数 Sample size of testing images
大型昆虫 Big insects	大螟 Sesamia inferens	背面Back image	330	140
	大螟 Sesamia inferens	腹面Abdomen image	428	140
	稻纵卷叶螟 Cnaphalocrocis medinalis	背面Back image	355	140
	稻纵卷叶螟 Cnaphalocrocis medinalis	腹面Abdomen image	364	140
	二化螟 Chilo suppressalis	背面Back image	355	140
	二化螟 Chilo suppressalis	腹面Abdomen image	364	140
	大型非目标昆虫Big non-target insects		5800	140
小型昆虫 Small insects	白背飞虱Sogatella furcifera	侧面Lateral image	3200	215
	褐飞虱Nilaparvata lugens	侧面Lateral image	800	215
	小型非目标昆虫Small non-target insects		6000	215

特征参数 Feature	大型目标害虫 Big pests		小型目标害虫 Small pests
特征参数 Feature	识别率 Identification rate	误检率 False detection rate	识别率 Identification rate	误检率 False detection rate
颜色 Color	52.9	18.4	87.7	14.9
形态 Shape	75.9	15.6	84.2	19.7
纹理 Texture	45.7	28.6	72.3	17.9
颜色+形态 Color+Shape	88.6	11.0	93.1	6.9
颜色+纹理 Color+Texture	73.8	13.2	91.6	7.3
形态+纹理 Shape+Texture	83.6	12.1	90.2	6.5
颜色+形态+纹理 Color+Shape+Texture	90.5	8.4	94.9	4.9

特征参数 Feature	大型目标害虫 Big pests		小型目标害虫 Small pests
特征参数 Feature	识别率 Identification rate	误检率 False detection rate	识别率 Identification rate	误检率 False detection rate
颜色 Color	52.9	18.4	87.7	14.9
形态 Shape	75.9	15.6	84.2	19.7
纹理 Texture	45.7	28.6	72.3	17.9
颜色+形态 Color+Shape	88.6	11.0	93.1	6.9
颜色+纹理 Color+Texture	73.8	13.2	91.6	7.3
形态+纹理 Shape+Texture	83.6	12.1	90.2	6.5
颜色+形态+纹理 Color+Shape+Texture	90.5	8.4	94.9	4.9

特征参数 Feature	大型目标害虫 Big pests		小型目标害虫 Small pests
特征参数 Feature	识别率 Identification rate	误检率 False detection rate	识别率 Identification rate	误检率 False detection rate
HOG (Histograms of oriented gradients)	87.6	5.4	87.9	8.3
LBP (Local binary pattern)	23.1	9.7	63.3	17.3
Gabor	72.9	6.9	86.4	8.9
颜色+纹理+形态+HOG Color+Shape+Texture+HOG	91.4	8.6	94.2	5.4