Research On Image Classification Algorithm Based On Deep Learning

Images are the visual basics for people to perceive the environment. According to images, people can get important information from the outside world. So it is of great significance to make machines automatically complete the image recognition and classification. Feature extraction is the core part of image classification, and extracting effective features can increase the recognition accuracy of recognition system. It is very important to research efficient algorithm for extracting features in the field of image classification. Deep learning is multi-layer network architecture, it establishes and simulates the hierarchical structure of human brain, processing data and extracting features from low to high level of the image, speech, text and other input data. Therefore, deep learning has widely applied in the field of image classification. However, deep learning itself has some defects, such as spends too much time in the process of adjusting parameters, and prones to over fitting and so on. In view of these problems, the main research contents of this paper are as follows:Firstly, this paper studies the feasibility and significance of using Extreme Learning Machine (ELM) as the classifier for Convolutional Neural Network (CNN) and then proposes the hybrid deep model Convolutional Neural Network-Extreme Learning Machine (CNN-ELM). First step, training the original CNN until it is convergent. Then the last layer of CNN was replaced by ELM. Finally, ELM completes a new classification. CNN-ELM integrates the synergy of two classifiers, and experiments show that CNN-ELM improves the classification accuracy of CNN.Then, this paper studies the feasibility and significance of using deep architectures with random weights to solve the problem of long training time in deep learning. Extreme Learning Machine with Kernel (KELM) was proposed based on ELM, and introduced kernel function into ELM. It has a better classification effect than ELM. Then we propose a deep model with random weights based on KELM: Convolutional Extreme Learning Machine with Kernel (CKELM). In CKELM model, alternate convolutional layers and sub-sampling layers with random weights as hidden layers added to the original KELM. Experiments show that CKELM can not only guarantee the classification accuracy but also greatly shorten the training time of depth algorithm.Finally, studies the feasibility and significance of deep auto encoder algorithm based on DropConnect. DropConnect as a new regularization algorithm has showed outstanding performance in dealing with the problem of over fitting. So this paper proposes a deep auto encoder model based on DropConnect:DropConnect Deep Auto Encoder (DDAE). And experiments show that introducing DropConncet ideal into Auto Encoder can effectively improve the classification ability of deep model.