Improvement Of Kernel Extreme Learning Machine And Its Application In Tumor Recognition

Extreme Learning Machine(ELM)is essentially an improved single hidden layer feedforward neural network,in which the parameters between input layer and hidden layer are randomly generated,and the connection weights between hidden layer and output layer are calculated by least square method.ELM has the advantages of fast learning speed and good generalization performance.In recent years,ELM has been widely used in practice and achieved good results.However,in practice,ELM also has some shortcomings,such as unstable results caused by random parameters,poor feature processing function caused by shallow structure and so on.Kernel Extreme Learning Machine(KELM)is an improved model of ELM.The kernel function theory of Support Vector Machine is introduced into ELM,which improves the generalization performance and maintains the stability of ELM.Like ELM,KELM has made great progress in regression and prediction.On the one hand,KELM uses matrix operation in training and classification process,which undoubtedly increases time complexity;on the other hand,many data sets such as natural images are redundant images,but the structure of KELM is shallow,which is not enough to extract better features from this part of the data.Therefore,KELM is studied and improved in this paper.It is applied to pattern recognition and a set of cancer image recognition system is constructed.The main work of this paper is as follows.First of all,a Kernel Extreme Learning Machine Based on Normal Equation(NE-KELM)is proposed.This algorithm applies the theory of normal equation to KELM,calculates the output weight of KELM by using normal equation,reduces the number of matrix operations and time complexity in the process of recognition.The algorithm is simulated on some data sets.Compared with the results of other algorithms,NE-KELM achieves better experimental results.Then,an improved model of Kernel Extreme Learning Machine(KELM)is proposed based on Kernel Extreme Learning Machine and feature dimension reduction theory.Firstly,the dimension of the sample feature information is reduced,and the feature dimension is reduced by using the fusing algorithm of PCA and LDA fusion.Then KELM is used to classify the data.The method is compared and analyzed on different data sets,which proves its superiority and obtains better results.Finally,the algorithm model is constructed into a cancer image recognition system with a visual interface.This system mainly realizes the training of the algorithm model and the recognition and classification of gastric epithelial cancer cells.In this paper,KELM algorithm is researched,and two improved algorithms are proposed and applied to data classification and image recognition.The simulation results show that the two improved methods proposed in this paper improve the performance of extreme learning machine algorithm.