基于栈式稀疏自编码器的青光眼眼底图像识别研究
|
摘要
青光眼是一种常见的威胁视神经及视觉功能的眼病,其具有发病率高,难以察觉等特点。但是目前对青光眼的识别诊断方法还不是很完善,且识别方法复杂,识别率也不高。因此提出了一种基于栈式稀疏自编码器的眼底图像特征提取及图像识别的方法。该方法采用逐层贪婪训练法从无标签的数据集中学习到数据的内部特征,将学习到的特征作为softmax分类器的输入。然后利用带标签的数据通过反向传播算法对稀疏自编码器进行调优。仿真实验分析中,使用测试集数据对该方法进行验证,精确度可达89%,并且优于实验中的其他方法,对青光眼的识别具有一定的实用价值。
Glaucoma is a common eye disease that attacks optic nerve and visual function,and it has the characteristics ofhigh incidence,difficult to detect and so on. However,the current methods for the diagnosis of glaucoma are not mature,and therecognition methods are complex and the recognition rate is not high. Therefore a method about feature extraction of fundus imageand image recognition based on stacked sparse auto-encoder is proposed. The method uses greedy layer-wise training method tolearn the internal characteristics of data from unlabeled data sets,and uses the learning features as input to the softmax classifier.Then,the sparse encoder is tuned by back propagation algorithm using labeled data. In the simulation experiment,the method isvalidated by using the test set data. The accuracy is up to 89%,and it is better than other methods in the experiment. It is of practi-cal value for the recognition of glaucoma.
Glaucoma is a common eye disease that attacks optic nerve and visual function,and it has the characteristics ofhigh incidence,difficult to detect and so on. However,the current methods for the diagnosis of glaucoma are not mature,and therecognition methods are complex and the recognition rate is not high. Therefore a method about feature extraction of fundus imageand image recognition based on stacked sparse auto-encoder is proposed. The method uses greedy layer-wise training method tolearn the internal characteristics of data from unlabeled data sets,and uses the learning features as input to the softmax classifier.Then,the sparse encoder is tuned by back propagation algorithm using labeled data. In the simulation experiment,the method isvalidated by using the test set data. The accuracy is up to 89%,and it is better than other methods in the experiment. It is of practi-cal value for the recognition of glaucoma.
引文
[1]文晓霞,谭德文.青光眼视功能损害及药物治疗研究进展[J].国际眼科杂志,2010,10(1):97-99.WEN Xiaoxia,TAN Dewen. Research Advances in VisualImpairment and Drug Treatment of Glaucoma[J]. Interna-tional Journal of Ophthalmology,2010,10(1):97-99.
[2]马建民.浅谈青光眼的危害[N].中国医药报,2006-06-12(B08).MA Jianming. The Harm of Glaucoma[N].China MedicalJournal,2006-06-12(B08).
[3]陈新建,邱晓雯,倪植枢.慢性闭角型青光眼的危害[J].实用眼科杂志,1988(7):41-43.CHEN Jianxin,QIU Xiaowen,NI Zhishu. The Harm ofChronic Angle Closure Glaucoma[J]. Journal of PracticalOphthalmology,1988(7):41-43.
[4]戴毅,孙兴怀.从流行病学角度认识青光眼的特点与危害[J].中国眼耳鼻喉科杂志,2016,16(3):179-181.DAI Yi,SUN Xinghuai. To Understand the Characteristicsand Harm of Glaucoma From Epidemiological Perspective[J].Chinese Journal of Ophthalmology and Otorhinolaryn-gology,2016,16(3):179-181.
[5]姚畅.眼底图像分割方法的研究及其应用[D].北京:北京交通大学,2009.YAO Shu. Research and Application of Fundus Image Seg-mentation Method[D]. Beijing:Beijing Jiaotong Universi-ty,2009.
[6]干能强.青光眼眼底图像的自动识别与分析方法研究[D].成都:西南交通大学,2008.GAN Nengqiang. Automatic Recognition and Analysis ofGlaucoma Fundus Images[D]. Chengdu:Southwest JiaoTong University,2008.
[7]黄元康.基于图像分割的青光眼眼底图识别研究[D].昆明:云南大学,2016.HUANG Yuankang. Recognition of Fundus Images ofGlaucoma Based on Image Segmentation[D]. Kunming:Yunnan University,2016.
[8]张成刚,姜静清.一种稀疏降噪自编码神经网络研究[J].内蒙古民族大学学报(自然科学版),2016,31(1):21-25,93.ZHANG Chenggang,JIANG Jingqing. Study on SparseDe-noising Auto-Encoder Neural Network[J]. Journal ofInner Mongolia University for Nationalities,2016,31(01):21-25,93.
[9]巩萍,王姗姗,罗举建.基于稀疏自编码神经网络的肺结节特征提取及良恶性分类[J].医疗卫生装备,2015,36(12):7-10,14.GONG Ping,WANG Shanshan,LUO Jujian. Feature Ex-traction and Benign or Malignant Classification of LungNodules Based on Sparse Auto-encoder Neural Network[J]. Chinese Medical Equipment Journal,2015,36(12):7-10,14.
[10]王勇,赵俭辉,章登义,等.基于稀疏自编码深度神经网络的林火图像分类[J].计算机工程与应用,2014,50(24):173-177.WANG Yong,ZHAO Jianhui,ZHANG Dengyi,et al.Forest fire image classification based on deep neural net-work of sparse autoencoder[J]. Computer Engineeringand Applications,2014,50(24):173-177.
[11]杨辉华,雒志超,蒋淑洁,等.稀疏降噪自编码算法用于近红外光谱鉴别真假药的研究[J].光谱学与光谱分析,2016,36(9):2774-2779.YANG Huihua,LUO Zhichao,JIANG Shujie,et al.Sparse Denoising Autoencoder Application in Identifica-tion of Counterfeit Pharmaceutical[J]. Spectroscopy and Spectral Analysis,2016,36(9):2774-2779.
[12]王杰,贾育衡,赵昕.基于稀疏自编码器的烟叶成熟度分类[J].烟草科技,2014(9):18-22.WANG Jie,JIA Yuheng,ZHAO Xin. Tobacco Leaf Ma-turity Classification Based on Sparse Auto-encoder[J].Tobacco Science&Technology,2014(9):18-22.
[13]戴晓爱,郭守恒,任淯,等.基于堆栈式稀疏自编码器的高光谱影像分类[J].电子科技大学学报,2016,45(3):382-386.DAI Xiaoai,GUO Shouheng,REN Yu,et al. Hyper spec-tral Remote Sensing Image Chassification Using theStacked Sparse Autoencoder[J]. Journal of University ofElectronic Science and Technology of China,2016,45(3):382-386.
[14]Ju Y,Guo J,Liu S. A Deep Learning Method CombinedSparse Autoencoder with SVM[C]//International Confer-ence on Cyber-Enabled Distributed Computing andKnowledge Discovery. IEEE Computer Society,2015:257-260.
[15]汪海波,陈雁翔,李艳秋.基于主成分分析和Softmax回归模型的人脸识别方法[J].合肥工业大学学报(自然科学版),2015,38(6):759-763.WANG Haibo,CHEN Yanxiang,LI Yanqiu. Face recog-nition method based on principal component analysis andSoftmax regression model[J]. Journal of hefei universityof technology,2015,38(6):759-763.
[2]马建民.浅谈青光眼的危害[N].中国医药报,2006-06-12(B08).MA Jianming. The Harm of Glaucoma[N].China MedicalJournal,2006-06-12(B08).
[3]陈新建,邱晓雯,倪植枢.慢性闭角型青光眼的危害[J].实用眼科杂志,1988(7):41-43.CHEN Jianxin,QIU Xiaowen,NI Zhishu. The Harm ofChronic Angle Closure Glaucoma[J]. Journal of PracticalOphthalmology,1988(7):41-43.
[4]戴毅,孙兴怀.从流行病学角度认识青光眼的特点与危害[J].中国眼耳鼻喉科杂志,2016,16(3):179-181.DAI Yi,SUN Xinghuai. To Understand the Characteristicsand Harm of Glaucoma From Epidemiological Perspective[J].Chinese Journal of Ophthalmology and Otorhinolaryn-gology,2016,16(3):179-181.
[5]姚畅.眼底图像分割方法的研究及其应用[D].北京:北京交通大学,2009.YAO Shu. Research and Application of Fundus Image Seg-mentation Method[D]. Beijing:Beijing Jiaotong Universi-ty,2009.
[6]干能强.青光眼眼底图像的自动识别与分析方法研究[D].成都:西南交通大学,2008.GAN Nengqiang. Automatic Recognition and Analysis ofGlaucoma Fundus Images[D]. Chengdu:Southwest JiaoTong University,2008.
[7]黄元康.基于图像分割的青光眼眼底图识别研究[D].昆明:云南大学,2016.HUANG Yuankang. Recognition of Fundus Images ofGlaucoma Based on Image Segmentation[D]. Kunming:Yunnan University,2016.
[8]张成刚,姜静清.一种稀疏降噪自编码神经网络研究[J].内蒙古民族大学学报(自然科学版),2016,31(1):21-25,93.ZHANG Chenggang,JIANG Jingqing. Study on SparseDe-noising Auto-Encoder Neural Network[J]. Journal ofInner Mongolia University for Nationalities,2016,31(01):21-25,93.
[9]巩萍,王姗姗,罗举建.基于稀疏自编码神经网络的肺结节特征提取及良恶性分类[J].医疗卫生装备,2015,36(12):7-10,14.GONG Ping,WANG Shanshan,LUO Jujian. Feature Ex-traction and Benign or Malignant Classification of LungNodules Based on Sparse Auto-encoder Neural Network[J]. Chinese Medical Equipment Journal,2015,36(12):7-10,14.
[10]王勇,赵俭辉,章登义,等.基于稀疏自编码深度神经网络的林火图像分类[J].计算机工程与应用,2014,50(24):173-177.WANG Yong,ZHAO Jianhui,ZHANG Dengyi,et al.Forest fire image classification based on deep neural net-work of sparse autoencoder[J]. Computer Engineeringand Applications,2014,50(24):173-177.
[11]杨辉华,雒志超,蒋淑洁,等.稀疏降噪自编码算法用于近红外光谱鉴别真假药的研究[J].光谱学与光谱分析,2016,36(9):2774-2779.YANG Huihua,LUO Zhichao,JIANG Shujie,et al.Sparse Denoising Autoencoder Application in Identifica-tion of Counterfeit Pharmaceutical[J]. Spectroscopy and Spectral Analysis,2016,36(9):2774-2779.
[12]王杰,贾育衡,赵昕.基于稀疏自编码器的烟叶成熟度分类[J].烟草科技,2014(9):18-22.WANG Jie,JIA Yuheng,ZHAO Xin. Tobacco Leaf Ma-turity Classification Based on Sparse Auto-encoder[J].Tobacco Science&Technology,2014(9):18-22.
[13]戴晓爱,郭守恒,任淯,等.基于堆栈式稀疏自编码器的高光谱影像分类[J].电子科技大学学报,2016,45(3):382-386.DAI Xiaoai,GUO Shouheng,REN Yu,et al. Hyper spec-tral Remote Sensing Image Chassification Using theStacked Sparse Autoencoder[J]. Journal of University ofElectronic Science and Technology of China,2016,45(3):382-386.
[14]Ju Y,Guo J,Liu S. A Deep Learning Method CombinedSparse Autoencoder with SVM[C]//International Confer-ence on Cyber-Enabled Distributed Computing andKnowledge Discovery. IEEE Computer Society,2015:257-260.
[15]汪海波,陈雁翔,李艳秋.基于主成分分析和Softmax回归模型的人脸识别方法[J].合肥工业大学学报(自然科学版),2015,38(6):759-763.WANG Haibo,CHEN Yanxiang,LI Yanqiu. Face recog-nition method based on principal component analysis andSoftmax regression model[J]. Journal of hefei universityof technology,2015,38(6):759-763.