Research On Kernel Extreme Learning Machine Algorithms And Application

In recent years,kernel extreme learning machine(KELM)has attracted more and more attention.The KELM is a single-hidden layer feedforward neural networks(SLFNs).Only the number of hidden layer nodes is needed to be set,and the least square method is adopted to calculate the weights.Therefore,the KELM has strong learning speed and generalization ability.Based on the theory of the KELM,a new combined KELM model is proposed in this paper,and an optimization algorithm is adopted to optimize the parameters of the combined KELM.Finally,the proposed model is applied to remote sensing satellite image recognition.The main works of this paper include:First of all,compared with the traditional BP neural network which uses the gradient descent method,the KELM can calculate unique optimal solution by using the least square method.Secondly,by combining the reproducing kernel function and the polynomial kernel function,a new combined KELM model is proposed.It is strictly proved that the reproducing kernel function can be used as an allowed extreme learning machine kernel function.Aim to the problem of many parameters of the model,the cuckoo search algorithm is adopted.Simulation results on bi-spiral benchmark and UCI datasets show that the proposed model has better generalization and robustness than the traditional classifier model.Finally,the combined KELM model is applied to remote sensing satellite image recognition to construct a classifier.The sparse coding algorithm is used to extract the image features and the images are classified based on the remote sensing satellite datasets of UC Merced Land Use.The results show that the proposed model can identify remote sensing satellite images with high accuracy.