Please wait a minute...
Journal of Zhejiang University (Agriculture and Life Sciences)  2019, Vol. 45 Issue (1): 126-134    DOI: 10.3785/j.issn.1008-9209.2017.09.043
Agricultural engineering     
Detection method of slight bruises of apples based on hyperspectral imaging and RELIEF-extreme learning machine
Meng ZHANG1,2(),Guanghui LI1,2()
1. School of IoT Engineering, Jiangnan University, Wuxi 214000, Jiangsu, China
2. Engineering Research Center of IoT Technology Applications of Ministry of Education, Wuxi 214000, Jiangsu, China
Download: HTML   HTML (   PDF(1948KB)
Export: BibTeX | EndNote (RIS)      

Abstract  

In order to realize the rapid and nondestructive recognition of slight bruises of apples, a hyperspectral imaging technique (400-1 000 nm) was used. Hyperspectral images of sound and different damage time of Fuji apples were collected, and the average spectral reflectance and entropy were extracted from the region of interest (ROI) of the image. All the samples were divided into training set and test set (2∶1). The characteristic wave- bands extracted based on the spectral average reflectance and entropy using RELIEF algorithm were 17, 30, 35, 51, 61, 66, 94 and 120, respectively. Then, based on full wavebands and characteristic wavebands, an extreme learning machine (ELM) model was built, as comparison with support vector machine (SVM) and K-mean algorithm. The results showed that the recognition accuracy of ELM model for the test set based on the full wave- bands was 94.44%, and the accuracy of the Re-ELM model based on the characteristic wavebands was 96.67%, and the accuracy of the Re-SVM and Re-K mean models for the characteristic wavebands were 92.22% and 91.67%, respectively, which demonstrated that the Re-ELM was a more effective method for the bruise apple classification. Subsequently, an apple damage detection algorithm based on the characteristic wavebands and image processing was proposed, which performed an independent component algorithm (ICA) transformation of the characteristic wavebands, and selected the third component image of the ICA transformation, and used adaptive threshold segmentation to obtain the bruise area on apples. The final detection accuracy of apple damage detection algorithm based on the image processing technology was over 94%, which indicates that the algorithm is efficient for identifying slight bruises of apples.



Key wordsapple bruise      hyperspectral imaging      nondestructive detection      extreme learning machine      independent component analysis     
Received: 04 September 2017      Published: 28 March 2019
CLC:  TP 391.4  
Corresponding Authors: Guanghui LI     E-mail: zhangmeng0110@126.com;ghli@jiangnan.edu.cn
Cite this article:

Meng ZHANG,Guanghui LI. Detection method of slight bruises of apples based on hyperspectral imaging and RELIEF-extreme learning machine. Journal of Zhejiang University (Agriculture and Life Sciences), 2019, 45(1): 126-134.

URL:

http://www.zjujournals.com/agr/10.3785/j.issn.1008-9209.2017.09.043     OR     http://www.zjujournals.com/agr/Y2019/V45/I1/126


基于RELIEF算法和极限学习机的苹果轻微损伤高光谱检测方法

采用高光谱成像技术(400~1 000 nm)对苹果轻微损伤进行快速识别及无损检测。采集苹果正常及不同损伤时间的高光谱图像,选择图像中合适的区域作为感兴趣区域并提取平均光谱反射率及图像熵信息,将采集的样本按2∶1的比例分为训练集和测试集。使用RELIEF算法基于光谱平均反射率及图像熵信息提取了8个特征波段(17、30、35、51、61、66、94和120),分别基于全波段和特征波段进行极限学习机(extreme learning machine, ELM)建模分析,并与支持向量机(support vector machine, SVM)和K-均值聚类算法进行比较。结果表明,基于全波段的ELM模型最终测试集识别率为94.44%,基于特征波段的RELIEF-极限学习机(Re-ELM)模型识别率为96.67%,基于特征波段的Re-SVM及Re-K均值模型的最终测试集识别率分别为92.22%和91.67%,证实了Re-ELM是一种更为有效的苹果损伤分类判别方法。在此基础上,基于图像处理技术和特征波段提出了一种苹果轻微损伤高光谱检测算法,使用该算法针对特征波段进行独立成分分析(independent component analysis, ICA)变换,选取ICA第3成分图像进行自适应阈值分割,从而获得损伤图像。对全部高光谱图像进行检测表明,该算法的最终识别率超过94%,说明该算法能够较为有效地识别苹果损伤区域。


关键词: 苹果损伤,  高光谱成像,  无损检测,  极限学习机,  独立成分分析 
Fig. 1 Average reflectance spectra of the sound and bruise regions on apple samples
Fig. 2 Average entropy of the sound and bruise regions on apple samples
Fig. 3 Weighting coefficient of average reflectance spectra of apple samples
Fig. 4 Weighting coefficient of average entropy of apple samples
Fig. 5 Discriminant results of training sets based on full wavebands

类型

Type

训练集Training set 测试集Test set

数量

Number

正确数

Correct No.

错误数

Incorrect No.

识别率

Accuracy/%

数量

Number

正确数

Correct No.

错误数

Incorrect No.

识别率

Accuracy/%

总计Total 360 353 7 98.06 180 170 10 94.44
S-1M 72 69 3 95.83 36 32 4 88.89
S-1D 72 71 1 98.61 36 34 2 94.44
S-2D 72 70 2 97.22 36 34 2 94.44
S-3D 72 71 1 98.61 36 35 1 97.22
S-4D 72 72 0 100.00 36 35 1 97.22
Table 1 Discriminant results of sound and bruised samples using ELM model and full wavebands

类型a)

Typea)

训练集Training set 测试集Test set

数量

Number

正确数

Correct No.

错误数

Incorrect No.

识别率

Accuracy/%

数量

Number

正确数

Correct No.

错误数

Incorrect No.

识别率

Accuracy/%

总计Total 360 355 5 98.61 180 174 6 96.67
S-1M 72 69 3 95.83 36 33 3 91.67
S-1D 72 71 1 98.61 36 34 2 94.44
S-2D 72 71 1 98.61 36 35 1 97.22
S-3D 72 72 0 100.00 36 36 0 100.00
S-4D 72 72 0 100.00 36 36 0 100.00
 

类型a)

Typea)

训练集Training set 测试集Test set

数量

Number

正确数

Correct No.

错误数

Incorrect No.

识别率

Accuracy/%

数量

Number

正确数

Correct No.

错误数

Incorrect No.

识别率

Accuracy/%

总计Total 360 352 8 97.78 180 166 14 92.22
S-1M 72 71 1 98.61 36 32 4 88.89
S-1D 72 70 2 97.22 36 32 4 88.89
S-2D 72 71 1 98.61 36 34 2 94.44
S-3D 72 70 2 97.22 36 33 3 91.67
S-4D 72 70 2 97.22 36 35 1 97.22
Table 3 Discriminant results of sound and bruised samples based on Re-SVM model

类型a)

Typea)

数量

Number

正确数

Correct No.

错误数

Incorrect No.

识别率

Accuracy/%

总计 Total 540 495 45 91.67
S-1M 108 84 24 77.78
S-1D 108 99 9 91.67
S-2D 108 106 2 98.15
S-3D 108 107 1 99.07
S-4D 108 99 9 91.67
 
Fig. 6 ICA transformation results of the bruise apples’ hyperspectral images based on the full (A) and characteristic wavebands (B)
Fig. 7 Flow chart of detection algorithm of slight bruise apple samples

样本

Sample

数量

Number

正确数

Correct

No.

错误数

Incorrect No.

识别率

Accuracy/%

总计Total 324 306 18 94.44
正常Sound 54 52 2 96.30

损伤

Bruise

1 min 54 49 5 90.74
1 d 54 50 4 92.59
2 d 54 50 4 92.59
3 d 54 52 2 96.30
4 d 54 53 1 98.15
Table 5 Detection results of slight bruise apple samples by image recognition
[1]   张保华,黄文倩,李江波,等 .基于高光谱成像技术和MNF检测苹果的轻微损伤.光谱学与光谱分析,2014,34(5):1367-1372.
ZHANG B H , HUANG W Q , LI J B , et al . Detection of slight bruises on apples based on hyperspectral imaging and MNF transform. Spectroscopy and Spectral Analysis, 2014,34(5):1367-1372. (in Chinese with English abstract)
[2]   田有文,程怡,王小奇,等 .基于高光谱成像的苹果虫伤缺陷与果梗/花萼识别方法.农业工程学报,2015,31(4):325-331.
TIAN Y W , CHENG Y , WANG X Q , et al . Recognition method of insect damage and stem/calyx on apple based on hyperspectral imaging. Transactions of the CSAE, 2015,31(4):325-331. (in Chinese with English abstract)
[3]   黄文倩,陈立平,李江波,等 .基于高光谱成像的苹果轻微损伤检测有效波长选取.农业工程学报,2013,29(1):272-277.
HUANG W Q , CHEN L P , LI J B , et al . Effective wavelengths determination for detection of slight bruises on apples based on hyperspectral imaging. Transactions of the CSAE, 2013,29(1):272-277. (in Chinese with English abstract)
[4]   程术希,邵咏妮,吴迪,等 .稻叶瘟染病程度的可见-近红外光谱检测方法.浙江大学学报(农业与生命科学版),2011,37(3):307-311.
CHENG S X , SHAO Y N , WU D , et al . Determination of rice leaf blast disease level based on visible-near-infrared spectroscopy. Journal of Zhejiang University (Agriculture and Life Sciences), 2011,37(3):307-311. (in Chinese with English abstract)
[5]   LEE W H, KIM M S, LEE H, et al . Hyperspectral near-infrared imaging for the detection of physical damages of pear. Journal of Food Engineering, 2014,130:1-7.
[6]   魏新华,吴姝,范晓冬,等 .基于高光谱成像分析的冬枣微观损伤识别.农业机械学报,2015,46(3):242-246.
WEI X H , WU S , FAN X D , et al . Identification of slight bruises on winter jujube based on hyperspectral imaging technology. Transactions of the Chinese Society for Agricultural Machinery, 2015,46(3):242-246. (in Chinese with English abstract)
[7]   BARANOWSKI P , MAZUREK W , PASTUSZKA-WO?NIAK J . Supervised classification of bruised apples with respect to the time after bruising on the basis of hyperspectral imaging data. Postharvest Biology and Technology, 2013,86:249-258.
[8]   SUN Y , GU X Z , SUN K , et al . Hyperspectral reflectance imaging combined with chemometrics and successive projections algorithm for chilling injury classification in peaches. LWT—Food Science and Technology, 2016,75:557-564.
[9]   刘思伽,田有文,张芳,等 .采用二次连续投影法和BP人工神经网络的寒富苹果病害高光谱图像无损检测.食品科学,2017,38(8):277-282.
LIU S J , TIAN Y W , ZHANG F , et al . Hyperspectral imaging for nondestructive detection of Hanfu apple diseases using successive projections algorithm and BP neural network. Food Science, 2017,38(8):277-282. (in Chinese with English abstract)
[10]   MIN H , WAN X M , MIN Z , et al . Detection of insect-damaged vegetable soybeans using hyperspectral transmittance image. Journal of Food Engineering, 2013,116:45-49.
[11]   ABBOTT J A , LU R F , UPCHURCH B L , et al . Technologies for nondestructive quality evaluation of fruits and vegetables//Horticultural Reviews . Hoboken, United States: John Wiley & Sons Inc., 1997:1-120.
[12]   RAJKUMAR P , WANG N , EIMASRY G , et al . Studies on banana fruit quality and maturity stages using hyperspectral imaging. Journal of Food Engineering, 2012,108(1):194-200.
[13]   朱启兵,冯朝丽,黄敏,等 .基于图像熵信息的玉米种子纯度高光谱图像识别.农业工程学报,2012,28(23):271-276.
ZHU Q B , FENG Z L , HUANG M , et al . Maize seed classification based on image entropy using hyperspectral imaging technology. Transactions of the CSAE, 2012,28(23):271-276. (in Chinese with English abstract)
[14]   张翔,邓赵红,王士同,等 .极大熵Relief特征加权.计算机研究与发展,2011,48(6):1038-1048.
ZHANG X , DENG Z H , WANG S T , et al . Maximum entropy relief feature weighting. Journal of Computer Research and Development, 2011,48(6):1038-1048. (in Chinese with English abstract)
[15]   HUANG G B , ZHU Q Y , SIEW C K . Extreme learning machine: theory and applications. Neurocomputing, 2006,70(1/2/3):489-501.
[16]   丁世飞,齐丙娟,谭红艳 .支持向量机理论与算法研究综述.电子科技大学学报,2011,40(1):2-10.
DING S F , QI B J , TAN H Y . An overview on theory and algorithm of support vector machines. Journal of University of Electronic Science and Technology of China, 2011,40(1):2-10. (in Chinese with English abstract)
[17]   段明秀 .层次聚类算法的研究及应用.长沙:中南大学,2009.
DUAN M X . Research and application of hierarchical clustering algorithm. Changsha: Central South University, 2009. (in Chinese with English abstract)
[1] GUI Jiangsheng, GU Min, WU Zixian, BAO Xiao’an. Detection of multiple pesticide residues on the surface of broccoli based on hyperspectral imaging[J]. Journal of Zhejiang University (Agriculture and Life Sciences), 2018, 44(5): 643-648.
[2] Cao Leping, Wen Zhiyuan. Citrus quality grading based on statistical complexity measurement and multifractal spectrum method[J]. Journal of Zhejiang University (Agriculture and Life Sciences), 2015, 41(03): 309-319.
[3] FENG Lei,ZHANG Derong, CHEN Shuangshuang,FENG Bin, XIE Chuanqi,CHEN Youyuan,HE Yong. Early detection of gray mold on eggplant leaves using hyperspectral imaging technique.[J]. Journal of Zhejiang University (Agriculture and Life Sciences), 2012, 38(3): 311-317.