Research On A New Algorithm Of Image Classification Based On Extreme Learning Machine

Image classification is not only in-depth development of pattern recognition but also one of the key techniques in image processing. At the beginning of the classification process, it make analysis on the information of image by using computer, and finish image features’extraction and selection which could be used for determining the argument of mathematical model used in the training phase. Computer could finish classification for unkown images by using mathematical model finally. Currently, neural networks technology is the most widely applied in digital image processing, artificial neural networks—ANNs has increasingly become a very important tool in image classification with the development of its theory. However, traditional ANNs has some defects which are the impediment to its development For this reason, this paper imported the newest theory achievement-- extreme learning machine(ELM) inorder to overcome the defects of traditional ANNs’ learning algorithm.Further research was done based on image classification and ELM in this paper and main study contents are as follows:(1) Firstly, some problems exist in the classification process and classification process were introduced systematically.Classification process, including image preprocessing method, feature extract and select and data processing and classification techniques were introduced in detail. Secondly, this paper improve LBP according to the image’s size and a new method called feature extraction of image texture based on improved LBP and DLA was proposed at the same time. The experimental results show that the features through this method have both the global and local characteristics of the texture, and, furthermore, they could not only reduce their dimension but also keep the image of the main information. At the end, the result of image classification would be better because of this features.(2) ELM has some defects in practical application. Inorder to overcome them, this paper proposed the Modified Online Sequential Extreme Learning Machine based on ELM. This new algorithm keep all the good bits of ELM and adopt new method for solving the weights of the output to improve the learning speed.Regularization factor and online Sequential learning were put into for improving robustness and flexibility at the last. All the improvement result in good performance of ELM.(3) Experiments based on the improved method and algorithm for image classification were done and the performance of ELM based on many different activation function was arrived by experiments. This paper made comparative experiment with the traditional ELM, OS-ELM, SVM, and analysis the classified results in detail. All the results shows that this new algorithm and the improvement is valid. The results indicate that the improved new algorithm has better performance and fast learning speed and more ideal results of classification could be get by using this new algorithm.