Research On Ensemble Learning Algorithm Based On Sparse Representation Residual Reconstruction

Machine learning technology plays an important role in the field of IT,especially for massive data processing.Machine learning techniques evolve classifiers to improve the ability to classify new data by learning from existing data.Sparse representation and ensemble learning can effectively solve the problem of image classification.The sparse representation classifier is usually effectively classified by calculating the l2 norm of the reconstructed residuals of the original data.In complex cases,the discrimination error of the classifier may be due to the fact that the difference of the l2 norm of the residual vectors generated by various types is not obvious.Ensemble learning classifier has the advantages of high classification accuracy and stable classification performance.The classification performance of ensemble classifier depends not only on the base classifier,but also on the training sample set.In this paper,we propose a sparse representation model for reconstructing residual errors,and the classification method of multiple classifiers is studied by using residual as the feature of classification.Using sparse representation to reconstruct residual error effectively solves the instability of sparse signal representation.The sparse representation of the signal and the reconstructed residual set are used as the input samples of the ensemble learning,which improves the classification accuracy of the ensemble classifier.Simulation results using MNIST database show that the correctness of the classification using residual features is better than other methods,and it can identify some of the occluded images very well.In addition,it has a better classification effect on small training samples.Sparsity of sparse representation is closely related to dictionary learning.Generally,complete dictionary is not suitable for sparse representation of samples.In this paper,the KSVD algorithm is added to the over complete dictionary to achieve the optimization of the classification of sparse representation residual residuals.KSVD algorithm is a mainstream dictionary optimization algorithm.Experiments show that the performance of classification is further improved by adding KSVD algorithm in the generation of sparse representation dictionary.