Fuzzy Rough C-Mean Based CNN Clustering And Classification For Image Big Data

Deep learning(DL)has proven to be a powerful paradigm for clustering and classification of the large-scale image dataset,and it indicates incredible possibilities for unsupervised and semi-supervised learning of representation with clustering and classification algorithms.The forms of Convolution Neural Networks are now state-of-the-art for many clustering and classification tasks.However,with the perpetual incrementation of digital images,there exist more and more redundant,irrelevant,and noisy samples which cause CNN running to gradually decrease,and CNN requires a large number of labeled samples for training the network.However,the labeled data is often difficult,expensive or time-consuming.To conquer these issues,we proposed an effective unsupervised and semi-supervised methods for a large-scale image dataset,which combines CNN with Fuzzy and Rough theory(fuzziness).The main idea behind this is to reduce the uncertainty in terms of vagueness and indiscernibility Roughness and removing noise samples by using CNN architecture from raw data more specifically.We proposed that novel algorithms as follows.1.First,a novel unsupervised clustering algorithms(FRCNN)is proposed in the dissertation.The main idea is that a high-level representation learned by multi-layers of CNN model,with one clustering layer,produce the initial cluster center.Then during training image clusters centers and representations are updating jointly.FRCM is utilized to update the cluster centers in the forward pass,while the parameters of proposed CNN are updated by the backward pass based on Stochastic Gradient Descent(SGD).2.Second,a Semi-supervised Fuzzy Rough Convolution Neural Network(SSFRCNN)approach is also proposed in this dissertation,to fused Fuzzy-Rough C-Mean clustering with Convolutional Neural Network(CNNs)to knowledge learned from simultaneously intra-model and inter-model relationships to form the final data representation to be classified,which can be achieved better performance.The main idea behind this is to reduce the uncertainty in terms of vagueness and indiscernibility by using Fuzzy-Rough C-Mean clustering and removing noise samples by using CNN from raw data more specifically.In the framework of our proposed semi-supervised approach,we use all the training data with abundant unlabeled data with few labeled data to train SSFRCNN model.3.Third,we propose a 2D-reduction algorithm as data pre-processing technique for classification models,e.g.(CNN,Easy Ensemble,so on).The main idea behind this is the prediction performance of the models highly depends on the quality of dataset.2D-reduction implies that the reduction of data in ways:feature reduction and noisy sample reduction.In the feature reduction stage,removing the set of irrelevant features by using rough set approach and for noisy sample reduction,we proposed a novel Rough-KNN Noise reduction algorithm for reduction of noisy samples.The Rough set could adequately select the noisy boundary samples to eliminate on the basis of KNN rules with no information loss,whose classes have been mislabeled.It is shown that proposed methods be more effectiveness with a number of experiments on benchmark large-scale image datasets and also compare with state-of-the-art supervised,unsupervised and semi-supervised learning models for image clustering and classification.Comprehensive experiments indicate that our proposed approaches show outstanding performance with significance.