Research And Application Based On ELM-LRF

With the rapid development of computer,image processing,pattern recognition and machine vision are used more and more widely.Image classification is one of the basic problems in the field of computer vision and pattern recognition.In the real world,we use human visual cognition for the outside world.Visual information is captured by the human eye from the outside,and then these information sent to the brain,even the appearance of an object by light,angle and size and other factors have effect on it,which also have the brain information in their classification.Accordingly,in computer vision,camera or other image equipment instead of human eyes,the brain with computer simulation,by processing and analyzing of camera data acquisition,so as to obtain the correct image content.Image classification is to explain the object or scene contains all of the images,the original image pixels for converting image characteristics to construct the description of image information,according to the classification of the image information obtained.In recent years,the technology has been widely used in robot vision and tactile recognition,which greatly improves the ability of robot to recognize objects.As one of the most important means for the robot to obtain environmental information,machine vision can increase the autonomy of the robot and improve the flexibility of the robot.People hope that the intelligent robot can perceive the surrounding environment and things such as information and quickly identify in the immediate things and the surface of the object attribute perception by touching objects,so as to better serve the people.A typical robot identification method is generally the introduction of deep neural network structure,such as CNNs,although this method is very good but time-consuming and inefficient.In recent years,Huang Guangbin et al.Proposed a new type of single layer neural network structure extreme learning machine(ELM),which greatly reduced the training time and improved the classification accuracy.On the basis of ELM,Huang Guangbin et al have introduced local receptive fields of local receptive fields based on extreme learning machine,the model input and connection between the hidden layer are sparse,and the corresponding feelings by the local wild(on the continuous probability distribution of sampling)surrounded by experiments show that the model can not only shorten the training time,but also improve the stability of the algorithm.Therefore,how to effectively improve the accuracy of image classification is an important research content in this field,which is of great theoretical significance.This paper focuses on the research of robot robust recognition,with ELM-LRF as the core model,the algorithm is applied to robot target recognition,which greatly improves the recognition efficiency and greatly shorten the operation time.The contributions of this work are summarized as follows:(1).We propose an architecture — multi-modal ELM-LRF frame work,to construct the nonlinear representation from different aspects of information sources.The important merit of such a method is that the training time is greatly shortened and the testing efficacy is highly improved.(2).Considering the high level characteristics and the complexity of the target image,the single ELM-LRF has been unable to meet the demand of massive image classification processing,therefore,we extend the single ELM-LRF into multilayer layers,that is,multiple convolution and pool operation.The depth neural network structure can not only extract the abstract information of the depth image,but also ensure the displacement invariance of the data feature attributes.(3).A neural network model to extract the expression level based on the characteristics of the data,the model uses rough set theory attribute reduction,with its ability to extract useful features.It selects the features of the above model,then set them as the input to HELM-LRF for feature extraction and classification.