| Most of the classification and regression techniques have been applied to image classification,object recognition,and information security.The ensemble learning can usefully increase the regression and classification effects of the model through a reasonable combination of multiple learners.In this paper,aiming at the single kernel cannot effectively improve the generalization ability of the learner,this paper studies the related methods and technologies that fuse multiple kernel learning and ensemble learning to improve the generalization and robustness of the regression model.The main work of this paper is as follows:(1)Proposing the Sparse Loss induced Kernel Ensemble Regression.Aiming at the problem that the single kernel cannot effectively improve the generalization ability of the basic learner,this paper studies the ensemble learning of multiple kernel regressors using ensemble learning to solve the problem of kernel function and its parameter selection in the single kernel regression.The adopted multiple kernel ensemble model establishes a sparse loss of the multiple kernel based regressor in multiple Hilbert spaces,so that the model is jointly optimized and corresponding weights are obtained throughout the training process,and finally the multiple kernel reduction ensemble is established regression model.Experiments are performed on regression and classification tests on the UCI public data set and computer image data set respectively.The experimental results show that this algorithm has achieved better regression and classification results than other ensemble algorithms such as Random Forest,Adaboost,XGBoost and other algorithms.(2)Proposing the Image Classification Method based on Multiple Neural Network Ensemble.The thesis further studies an image classification method based on the multiple deep neural network ensemble.By combining ensemble learning with multiple activation functions,the traditional neural network is extended to an ensemble network structure with multiple neural networks in parallel,so that the neural network at each layer of the structure can store valid feature information and thus improve the model generalization capability.In MNIST,the CIFAR-10 computer image dataset was tested for classification and the experimental results showed that the algorithm can achieve better classification accuracy compared to convolutional neural networks and other parallel network algorithms.(3)Designing an image classification prototype system based on multiple kernel ensemble regression.Based on the above research,a prototype image classification system based on multiple kernel ensemble regression is designed.The prototype system is divided into a multi-user role module and an image classification module.The image classification results show that the image recognition system can achieve the required functions. |
