基于CNN与RoELM的图像分类算法研究
|
摘要
针对极限学习机(ELM)等单隐层前馈神经网络(SLFN)算法在特征提取方面不精准以及深度学习(DL)算法学习时间长,易陷入局部最小值等问题,提出了一种卷积混合模型极限学习机(CNN-RoELM)。该算法使用卷积神经网络(CNN)提取特征,通过把多个卷积层与降采样层作为隐层来实现图像特征提取,并采用随机权值,从而极大地减少了提取特征过程中的时间;然后利用加权最小二乘法来计算鲁棒极限学习机(RoELM)的输出权值,实现图像快速分类。实验结果表明,该算法能够高效,精准地实现图像分类,同时又保留了较好的鲁棒性和泛化性。
The single-hidden layer feedforward neural network(SLFN)algorithm,such as ELM(extreme learning machine),is not accurate in feature extraction.Deep learning algorithm(DL)needs to spend much time in the process of adjusting parametersand is easy to get into the local minimum value. Focused on the problems,a kind of algorithm,CNN-RoELM,based on CNN and RoELM is proposed. The algorithm uses the convolution neural network(CNN)to extract the feature. The image feature extraction is realized by using multiple convolutions and descending layers as hidden layers,and the random weights are used,which greatly reduces the time in the process of extracting features. Then the weighted least squares method is used to calculate the output weight ofthe robust limit learning machine(RoELM),and the image classification is realized quickly. The theoretical analysis and simulation results show that the algorithm can achieve image classification quickly and accurately,while preserving good robustness andgeneralization.
The single-hidden layer feedforward neural network(SLFN)algorithm,such as ELM(extreme learning machine),is not accurate in feature extraction.Deep learning algorithm(DL)needs to spend much time in the process of adjusting parametersand is easy to get into the local minimum value. Focused on the problems,a kind of algorithm,CNN-RoELM,based on CNN and RoELM is proposed. The algorithm uses the convolution neural network(CNN)to extract the feature. The image feature extraction is realized by using multiple convolutions and descending layers as hidden layers,and the random weights are used,which greatly reduces the time in the process of extracting features. Then the weighted least squares method is used to calculate the output weight ofthe robust limit learning machine(RoELM),and the image classification is realized quickly. The theoretical analysis and simulation results show that the algorithm can achieve image classification quickly and accurately,while preserving good robustness andgeneralization.
引文
[1]Huang G B,Zhu Q Y,Siew C K. Extreme learning machine:a new learning scheme of feed forward neural networks[C]//IEEE International Joint Conference on NeuralNetworks,2005,2(2):985-990.
[2]Vapnik VN. The nature of statistical learning theory[M].Springer-Verlag,1995,8(6):988-999.
[3]DENG Wan-Yu,ZHENG Qing-Hua,CHEN Lin,et al. Research on Extreme Learning of Neural Networks[J]. Chinese Journal of Computers,2010,33(2):279-287.
[4]Huang G B,Zhou H M,Ding X J,et al. Extreme learningmachine for regression and multiclass classification[J].IEEE Transactions on Systems Man&Cybernetics Part B,2012,42(2):513-529.
[5]Jarrett K,Kavukcuoglu K,Ranzato M,et al. What is thebest multi-stage architecture for object recognition[C]//IEEE International Conference on Computer Vision,2010,30(2):2146-2153.
[6]Y Miche,A Sorjamaa,P Bas,et al. OP-ELM:OptimallyPruned Extreme Learning Machine[J]. IEEE Transactionson Neural Networks,2010,21(1):158-162.
[7] Kai Zhang,MinxiaLuo. Outlier-robust extreme learningmachine for regression problems[J]. Neurocomputing,2015,151(3):1519-1527.
[8]Yongping Zhao,Kangkang Wang. Fast cross validation forregularized extreme learning machine[J]. Journal of Systems Engineering and Electronics,2014,25(5):895-900.
[9]吴耀华.线性模型中M估计的渐进性质[J].应用数学学报,1994,7(yy):401-410.WU Yaohua. Asymptotic behavior of M-estimators in linear models[J]. Acta Mathematicae Applicatae Sinica,1994,7(yy):401-410.
[10]James O,Street,Raymond J,Carroll,David Ruppert. Anote on computing robust regression estimates via iteratively reweighted least squares[J]. The American Statistician,1988,42(2):152-154.
[11]张静,陈益强,纪雯,等.基于CNN与ELM的二次超分辨率重构方法研究[J].计算机学报,2017,40(52):1-17.ZHANG Jing,CHEN Yiqiang,JI Wen,et al. Two-tie image super-resolution based on CNN and ELM[J]. Chinese Journal of Computers,2017,40(52):1-17.
[12]乔俊飞,李凡军,杨翠丽.随机权神经网络研究现状与展望[J].智能系统学报,2016,11(6):758-767.QIAO Junfei,LI Fanjun,YANG Cuili. Review and prospect on neural networks with random weights[J].CAAITransactions on Intelligent Systems,2016,11(6):758-767.
[13]Shen F,Wang L,Zhang J. Reduced extreme learning machine employing SVM technique[J]. Journal of Huazhong University of Science&Technology,2014,42(6):107-111.
[14]Casta A O,F Ndez-Navarro,s-Mart,et al. PCA-ELM:ARobust and Pruned Extreme Learning Machine ApproachBased on Principal Component Analysis[J]. Kluwer Academic Publishers,2013,37(3):377-392.
[15]Zhang Y,Jin J,Wang X,et al. Motor imagery EEG classification via Bayesian extreme learning machine[C]//International Conference on Informati on Science and Technology,2016:27-30.
[16]Mohamed A,Dahl G E,Hinton G. Acoustic Modeling Using Deep Belief Networks[J]. IEEE Transactions on Audio Speech and Language Processing,2012,20(1):14-22.
[2]Vapnik VN. The nature of statistical learning theory[M].Springer-Verlag,1995,8(6):988-999.
[3]DENG Wan-Yu,ZHENG Qing-Hua,CHEN Lin,et al. Research on Extreme Learning of Neural Networks[J]. Chinese Journal of Computers,2010,33(2):279-287.
[4]Huang G B,Zhou H M,Ding X J,et al. Extreme learningmachine for regression and multiclass classification[J].IEEE Transactions on Systems Man&Cybernetics Part B,2012,42(2):513-529.
[5]Jarrett K,Kavukcuoglu K,Ranzato M,et al. What is thebest multi-stage architecture for object recognition[C]//IEEE International Conference on Computer Vision,2010,30(2):2146-2153.
[6]Y Miche,A Sorjamaa,P Bas,et al. OP-ELM:OptimallyPruned Extreme Learning Machine[J]. IEEE Transactionson Neural Networks,2010,21(1):158-162.
[7] Kai Zhang,MinxiaLuo. Outlier-robust extreme learningmachine for regression problems[J]. Neurocomputing,2015,151(3):1519-1527.
[8]Yongping Zhao,Kangkang Wang. Fast cross validation forregularized extreme learning machine[J]. Journal of Systems Engineering and Electronics,2014,25(5):895-900.
[9]吴耀华.线性模型中M估计的渐进性质[J].应用数学学报,1994,7(yy):401-410.WU Yaohua. Asymptotic behavior of M-estimators in linear models[J]. Acta Mathematicae Applicatae Sinica,1994,7(yy):401-410.
[10]James O,Street,Raymond J,Carroll,David Ruppert. Anote on computing robust regression estimates via iteratively reweighted least squares[J]. The American Statistician,1988,42(2):152-154.
[11]张静,陈益强,纪雯,等.基于CNN与ELM的二次超分辨率重构方法研究[J].计算机学报,2017,40(52):1-17.ZHANG Jing,CHEN Yiqiang,JI Wen,et al. Two-tie image super-resolution based on CNN and ELM[J]. Chinese Journal of Computers,2017,40(52):1-17.
[12]乔俊飞,李凡军,杨翠丽.随机权神经网络研究现状与展望[J].智能系统学报,2016,11(6):758-767.QIAO Junfei,LI Fanjun,YANG Cuili. Review and prospect on neural networks with random weights[J].CAAITransactions on Intelligent Systems,2016,11(6):758-767.
[13]Shen F,Wang L,Zhang J. Reduced extreme learning machine employing SVM technique[J]. Journal of Huazhong University of Science&Technology,2014,42(6):107-111.
[14]Casta A O,F Ndez-Navarro,s-Mart,et al. PCA-ELM:ARobust and Pruned Extreme Learning Machine ApproachBased on Principal Component Analysis[J]. Kluwer Academic Publishers,2013,37(3):377-392.
[15]Zhang Y,Jin J,Wang X,et al. Motor imagery EEG classification via Bayesian extreme learning machine[C]//International Conference on Informati on Science and Technology,2016:27-30.
[16]Mohamed A,Dahl G E,Hinton G. Acoustic Modeling Using Deep Belief Networks[J]. IEEE Transactions on Audio Speech and Language Processing,2012,20(1):14-22.