| Broad learning system is a very fast and effective discriminant learning which developed by C.L.P.Chen,Z.Liu and others.It is a very fast and effective discriminant learning,which avoids the disadvantages of complex network structure,large number of parameters and large computation in deep learning neural networks.However,since the development of BLS is still in its infancy,the model needs further analysis,improvement and verification.Therefore,based on the broad learning system,this thesis studies the problem of its approximation and its application on face classification.First,the proof of the approximation theorem of continuous function defined on a compact set by broad learning system is given.The mathematical form of the output nk mq function of BLS is(?)In this thesis,it is proved that if the activation function of the enhancement node of BLS is not polynomial,for any continuous function f(x)∈C(K)defined on the compact set K,there is(?)that is ?ε>0,?nk∈N,mq∈N’ and parameter set w,so that ‖f(x)-fw(X)‖22<ε.The approximation conclusion of the probability expectation of BLS on the measurable set is a corollary of this method.Second,a broad learning system algorithm based on convolution neural networks processing is proposed and is applied to face classification.CNN and BLS are combined,principal component analysis is used to learn the mapping matrix,convolution and pooling operations are used to extract enhancement nodes from the input image.Then,principal component analysis is performed on the merged feature map,where the enhanced features are extracted.Therefore,BLS is reconstructed,and CNNBLS algorithm is proposed.The face classification numerical experiments are done on the ORL and Yale databases.The experimental results show that BLS has a higher calculation accuracy,fewer features calculations than traditional principal component methods.The generalization ability of BLS is also very well.Finally,the study of the classification of faces by semi-supervised width networks on regular popularity is explored.BLS is extended based on semi-supervised learning of manifold regularization framework.The semi-supervised broad learning system(SS-BLS)and its algorithm are proposed.Then,SS-BLS and convolution function are combined to establish SS-CBLS,the test accuracy and training time of SS-CBLS and traditional convolutional neural network on face classification are compared by ORL and Yale face database respectively.The experimental results show that compared with traditional CNN,SS-CBLS has higher classification accuracy,faster operation speed and stronger generalization ability for different data on face classification numerical experiments. |
