[1]许伟栋,赵忠盖.基于卷积神经网络和支持向量机算法的马铃薯表面缺陷检测[J].江苏农业学报,2018,(06):1378-1385.[doi:doi:10.3969/j.issn.1000-4440.2018.06.025]
　XU Wei-dong,ZHAO Zhong-gai.Potato surface defects detection based on convolution neural networks and support vector machine algorithm[J].,2018,(06):1378-1385.[doi:doi:10.3969/j.issn.1000-4440.2018.06.025]
点击复制

基于卷积神经网络和支持向量机算法的马铃薯表面缺陷检测()
分享到：

江苏农业学报[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2018年06期
页码:
1378-1385
栏目:
加工贮藏·质量安全
出版日期:
2018-12-25

文章信息/Info

Title:
Potato surface defects detection based on convolution neural networks and support vector machine algorithm
作者:
许伟栋; 赵忠盖
(江南大学轻工过程先进控制教育部重点实验室，江苏无锡214000)
Author(s):
XU Wei-dong; ZHAO Zhong-gai
(Key Laboratory of Advanced Control for Light Industry Processes, Ministry of Education, Jiangnan University, Wuxi 214000, China)
关键词:
表面缺陷; 马铃薯; 机器视觉; 卷积神经网络; 支持向量机
Keywords:
surface defect; potato; machine vision; convolution neural network; support vector machine
分类号:
TP391.41;S532
DOI:
doi:10.3969/j.issn.1000-4440.2018.06.025
文献标志码:
A
摘要:
马铃薯表面缺陷检测是马铃薯分级的重要组成部分。传统的马铃薯表面缺陷检测方法通常涉及到手工特征提取和特征判断,但是马铃薯生长环境复杂，缺陷种类繁多，提取合适的特征是一个难题。为了解决上述问题，本试验提出一种基于改进的卷积神经网络(Convolution Neural Networks，CNN)和支持向量机(Support Vector Machine，SVM)模型的马铃薯表面缺陷检测新方法。该模型通过CNN自动提取马铃薯图片深度特征，利用特征向量训练SVM得到分类器。此外，改进的CNN中采用dropout正则化技术，能有效减小模型过拟合；加入1×1卷积层，加快模型运算时间。试验中，对CNN模型学习率和训练次数等超参数进行了对比选择，基于Adam优化算法通过GPU加速技术进行CNN模型训练；采用网格搜索算法对SVM参数进行优选。试验样本集由实验室机器视觉平台和数据增广方法所得图片组成。试验结果表明,本试验设计的CNN+SVM改进模型能解决现有研究中存在的问题，且性能优于常规CNN模型和传统检测方法，算法运行速度更快，准确率达99.20%。
Abstract:
Detection of potato surface defects is an important part of potato grading. Traditional potato surface defect detection methods usually involve manual feature extraction and judgement, in addition, the growth environment of potato is complex and the types of deficiencies are numerous, extracting appropriate features is a difficult problem. In order to solve the above problems, this paper presented a new method of potato surface defect detection based on improved convolution neural networks (CNN) and support vector machine (SVM) algorithm. The model automatically extracted the depth feature of potato picture through CNN, and the SVM was trained using eigenvector to get the classifier. The improved CNN model used the dropout regularization technique, which could effectively reduce the model overfitting. Moreover, 1×1 convolution layers were used to reduce the model operation time. In addition, the comparison and selection of hyperparameters such as learning rate and training times in the CNN model were made. The CNN+SVM model was trained with GPU acceleration technique based on Adam optimization algorithm. Moreover, grid search algorithm was used to optimize the parameters of SVM. The experimental sample set consisted of pictures obtained by the laboratory machine vision platform and the data augmentation method. The experimental results showed that the improved CNN+SVM model designed in this study could solve the existing problems in current research and its performance was superior to the typical CNN model and traditional detection methods. The CNN-SVM algorithms run faster and have a classification accuracy of 99.2% on experimental datasets.

参考文献/References:

[1]邓丽苗,杜宏伟,徐艳,等.基于机器视觉的马铃薯智能分选方法与实现[J].中国农机化学报,2015，36(5)：145-150.
[2]刘韦.基于机器视觉的马铃薯表面缺陷检测算法的研究[D].大庆：黑龙江八一农垦大学,2013.
[3]汪成龙,李小昱,武振中,等.基于流形学习算法的马铃薯机械损伤机器视觉检测方法[J].农业工程学报,2014，30(1):245-252.
[4]周竹,黄懿,李小昱,等.基于机器视觉的马铃薯自动分级方法[J].农业工程学报,2012, 28(7):178-183.
[5]HASSAN K R. Potato surface defect detection in machine vision system [J]. African Journal of Agricultural Research, 2012, 7(5)：844-850.
[6]李锦卫,廖桂平,金晶,等.基于灰度截留分割与十色模型的马铃薯表面缺陷检测方法[J].农业工程学报, 2010, 26(10):236-242.
[7]RAZ M N, MOUSAVI B S, SOLEYMANI F. A real-time mathematical computer method for potato inspection using machine vision [J]. Computers & Mathematics with Applications, 2012, 63(1): 268-279.
[8]BARNES M, DUCKETT T, CIELNIAK G, et al. Visual detection of blemishes in potatoes using minimalist boosted classifiers [J]. Journal of Food Engineering, 2010, 98(3):339-346.
[9]徐姗姗,刘应安,徐昇.基于卷积神经网络的木材缺陷识别[J]. 山东大学学报(工学版), 2013, 43(2):23-28.
[10]FAGHIH-ROOHJ S, HAJIZADEH S, NUNEZ A, et al. Deep convolutional neural networks for detection of rail surface defects: international joint conference on neural networks [C]Vancouver: IEEE, 2016.
[11]余永维,殷国富,殷鹰,等.基于深度学习网络的射线图像缺陷识别方法[J].仪器仪表学报, 2014, 35(9):2012-2019.
[12]刘明,李国军,郝华青,等.基于卷积神经网络的T波形态分类[J].自动化学报, 2016, 42(9):1339-1346.
[13]龚丁禧,曹长荣.基于卷积神经网络的植物叶片分类[J].计算机与现代化, 2014, 2014(4):12-15.
[14]KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks: International conference on neural information processing systems [C] Lake Tahoe：NIPS foundation,2012:1097-1105.
[15]ZEILER M D, FERGUS R. Visualizing and understanding convolutional networks: European conference on computer vision [C]. Amsterdam：IEEE , 2014:818-833.
[16]SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition: International conference on learning representations [C]. San Diego: IEEE,2015
[17]SZEGEDY C, LIU W, JIA Y, et al. Going deeper with convolutions: Computer vision and pattern recognition[C]. Boston：IEEE, 2015:1-9.
[18]SRIVASTAVA N, HINTON G, KRIIH EVEKYA, et al. Dropout: a simple way to prevent neural networks from overfitting [J] Journal of Machine Learning Research, 2014, 15（1）:1929-1958.
[19]HE K, ZHANG X, REN S, et al. Delving deep into rectifiers: surpassing human-level performance on imageNet classification: International conference on computer vision[C] Santiago：IEEE , 2015:1026-1034.
[20]KINGMA D P, BA J. Adam: A method for stochastic optimization: International conference on learning representations [C]. San Diego: IEEE 2015.
[21]FERREIRA A, GIRALDI G. Convolutional neural network approaches to granite tiles classification[J]. Expert Systems with Applications, 2017, 84(30):1-11.
[22]中华人民共和国农业部. 马铃薯等级规格:NY/T 1066-2006[S] 北京:中国农业出版社, 2006.
[23]郭跃东, 宋旭东. 梯度下降法的分析和改进[J]. 科技展望, 2016, 26(15):115-117.
[24]CHEN Y, JIANG H, LI C, et al. Deep feature extraction and classification of hyperspectral images based on convolutional neural networks[J]. IEEE Transactions on Geoscience & Remote Sensing, 2016, 54(10):6232-6251.

相似文献/References:

[1]何虎翼,谭冠宁,何新民,等.63 份马铃薯品种(系)资源农艺性状的主成分与聚类分析[J].江苏农业学报,2017,(01):27.[doi:10.3969/j.issn.1000-4440.2017.01.005 ]
　HE Hu-yi,TAN Guan-ning,HE Xin-min,et al.Principal component and cluster analysis for agronomic traits of 63 potato varieties or clones[J].,2017,(06):27.[doi:10.3969/j.issn.1000-4440.2017.01.005 ]
[2]徐玉伟,印敬明,白潇,等.马铃薯 StPYL1 和 StPYL8 基因的分子克隆与表达分析[J].江苏农业学报,2015,(01):23.[doi:10.3969/j.issn.1000-4440.2015.01.004]
　XU Yu-wei,YIN Jing-ming,BAI Xiao,et al.Molecular cloning and expression analysis of potato StPYL1 and StPYL8 genes[J].,2015,(06):23.[doi:10.3969/j.issn.1000-4440.2015.01.004]
[3]亢艳莉,申双和,张学艺,等.气候变化对宁夏南部山区马铃薯产量的影响及马铃薯水分供需特征分析[J].江苏农业学报,2017,(05):1056.[doi:doi:10.3969/j.issn.1000-4440.2017.05.015]
　KANG Yan-li,SHEN Shuang-he,ZHANG Xue-yi,et al.Effect of climate change on potato yield of Ningxia southern mountainous area and analysis of characteristics of water supply and demand in potato[J].,2017,(06):1056.[doi:doi:10.3969/j.issn.1000-4440.2017.05.015]
[4]王卓卓,何英彬,罗善军,等.基于冠层高光谱数据与马氏距离的马铃薯品种识别[J].江苏农业学报,2018,(05):1036.[doi:doi:10.3969/j.issn.1000-4440.2018.05.010]
　WANG Zhuo-zhuo,HE Ying-bin,LUO Shan-jun,et al.Variety identification of potatoes based on canopy hyperspectral data and Mahalanobis distance[J].,2018,(06):1036.[doi:doi:10.3969/j.issn.1000-4440.2018.05.010]
[5]黄强,郑顺林,郭函,等.氮增效剂对马铃薯叶片及土壤氮的影响[J].江苏农业学报,2019,(05):1087.[doi:doi:10.3969/j.issn.1000-4440.2019.05.013]
　HUANG Qiang,ZHENG Shun-lin,GUO Han,et al.Effects of nitrogen synergist on nitrogen in potato leaves and soil[J].,2019,(06):1087.[doi:doi:10.3969/j.issn.1000-4440.2019.05.013]
[6]许建民,颜志明,史培华,等.不同光谱及其组合对马铃薯干物质积累和分配的影响[J].江苏农业学报,2020,(01):32.[doi:doi:10.3969/j.issn.1000-4440.2020.01.005]
　XU Jian-min,YAN Zhi-ming,SHI Pei-hua,et al.Effects of different spectra and their combinations on dry matter accumulation and distribution in potato[J].,2020,(06):32.[doi:doi:10.3969/j.issn.1000-4440.2020.01.005]
[7]许建民,刘艳,颜志明,等.不同光谱对马铃薯种薯品质的影响[J].江苏农业学报,2020,(05):1105.[doi:doi:10.3969/j.issn.1000-4440.2020.05.005]
　XU Jian-min,LIU Yan,YAN Zhi-ming,et al.Effects of different spectra on quality of seed potato[J].,2020,(06):1105.[doi:doi:10.3969/j.issn.1000-4440.2020.05.005]
[8]杨茜,刘吉利,贺锦红,等.栽培模式对宁南地区马铃薯生理特性及产量的影响[J].江苏农业学报,2021,(03):555.[doi:doi:10.3969/j.issn.1000-4440.2021.03.002]
　YANG Qian,LIU Ji-li,HE Jin-hong,et al.Effects of cultivation pattern on physiological characteristics and yield of potatoes planted in southern Ningxia[J].,2021,(06):555.[doi:doi:10.3969/j.issn.1000-4440.2021.03.002]
[9]肖熙鸥,林文秋,陈卓,等.马铃薯抗青枯病育种研究进展[J].江苏农业学报,2021,(05):1344.[doi:doi:10.3969/j.issn.1000-4440.2021.05.033]
　XIAO Xi-ou,LIN Wen-qiu,CHEN Zhuo,et al.Research advances in potato breeding for bacterial wilt resistance[J].,2021,(06):1344.[doi:doi:10.3969/j.issn.1000-4440.2021.05.033]
[10]安珍,张茹艳,周春涛,等.铁肥对马铃薯生理特性、产量及品质的影响[J].江苏农业学报,2022,38(04):931.[doi:doi:10.3969/j.issn.1000-4440.2022.04.009]
　AN Zhen,ZHANG Ru-yan,ZHOU Chun-tao,et al.Effects of iron fertilizer on physiological characteristics, yield and quality of potato[J].,2022,38(06):931.[doi:doi:10.3969/j.issn.1000-4440.2022.04.009]

备注/Memo

备注/Memo:
收稿日期：2018-02-01 基金项目：国家自然科学基金项目（61573169）；江苏省六大人才高峰项目（2014-ZBZZ-010） 作者简介：许伟栋(1993-)，男，江苏无锡人，硕士研究生，主要研究方向为机器视觉农产品检测，（E-mail）1247621673@qq.com 通讯作者：赵忠盖，（E-mail）gaizihao@jiangnan.edu.cn
更新日期/Last Update: 2018-12-28