Information Extraction Of Remote Sensing Image Based On Compensatory Fuzzy Neural Network

Remote Sensing image is a real-time record of the characteristics of the electromagnetic spectrum of the ground,and it can reveal the object target and its distribution through the difference of the brightness value or the pixel value and the spatial change,displaying the true information of the shape,size,color and other information of the surface of the earth.In recent years,with the continuous development of sensor technology,various earth observation data from coarse to fine has been continuously produced,from which more useful information can be acquired and applied in national defense,agriculture,disaster monitoring,urban planning and other fields.Facing the massive remote sensing data with the scale of capacity growth at TB grade every day,the useful information that can be obtain in time is rather limited,and a large number of satellite remote sensing data can not be fully utilized.As a result,a great waste of remote sensing information resources has happened.The existing remote sensing image information extraction and massive data processing technology can not meet the requirements of the current remote sensing data application.It is an urgent problem that how to extract topographical information efficiently,accurately and automatically from the massive data in the remote sensing image.At present,the research on remote sensing image information extraction method is developing rapidly,mainly because it includes not only the contents of remote sensing technology,image processing and pattern recognition,but also the fields of land use,environmental monitoring,disaster prediction and urban planning.Due to the continuous development of the cutting edge fuzzy logic theory,fuzzy neural network theory,matter-element theory,gray theory system theory and rough set theory,as well as actual use of military and civilian,the remote sensing image information extraction method has been improved and developed continuously.Artificial Neural Network(ANN)has the characteristics of anti-interference,high fault tolerance,adaptive information processing and memory ability.In recent years,it has been successfully applied in the field of remote sensing image information extraction,and made a lot of research results.However,in the existing artificial neural network technology applied in the remote sensing image information extraction,the network training process is slow,and it is difficult to reach to the optimal value.At the same time,due to the target diversity,data uncertainty and other uncertain factors of remote sensing image,the remote sensing image classification accuracy has been affected,and the automatic extraction cannot be achieved,which has been unable to meet the urgent requirements of practical application.The combination of fuzzy theory and neural network can play its own advantages,and membership degree can describe the relationship between complex remote sensing information.Compensation fuzzy neural network is a hybrid system which combines compensation fuzzy logic and neural network,which consists of fuzzy control oriented and decision-oriented fuzzy neurons.The introduction of compensation fuzzy neural network enable the network to training from the correctly-defined fuzzy rules or the incorrectly-defined fuzzy rules,which result in the higher network fault tolerance,faster training and more stable system.Compensation fuzzy neural network can best embody the inherent characteristics of remote sensing image extraction,which is one of the important research topic of artificial intelligence and automatic extraction of information.At present,there is not much research on the extraction of remote sensing image information by fuzzy neural network,and there is no recognized mature technology in this field,so further research should be carried and improved.Therefore,the theme of this paper is:"remote sensing image information extraction based on compensated fuzzy neural network".This paper will introduce the fuzzy neural network theory into the remote sensing image information extraction,aiming to establish a suitable remote sensing image information extraction method for practical application,which provides a more effective scientific method and scientific basis for the practical application of remote sensing images in related fields.Based on the deep analysis and research of the existing remote sensing image information extraction method,this paper proposes a information extraction method of remote sensing image based on compensation fuzzy neural network,and conduct deep research in the key issue of information extraction by this method,including the network structure design,membership function determination,sample construction,and feature selection and extraction.The corresponding calculation model of compensation fuzzy neural network is established,and the method and model are studied in remote sensing image information extraction.In the experimental study,a deep research about the remote sensing image information extraction of single feature,dual feature combination,three-feature combination and multi-feature fusion is carried deeply.At the same time,in order to verify the feasibility and validity of the method of remote sensing image extraction based on compensation fuzzy neural network,the Object oriented information extraction method using the industry recognized software eCognition and the index method of NDTBI the author created is used to extract the remote sensing image information,and the comparative study was carried out.The main work and the main conclusions are as follows.The conclusion(1),(2),(3)and(8)are the innovation of this article.(1)Based on the analysis of existing remote sensing image information extraction method,a remote sensing image information extraction method based on compensated fuzzy neural network theory is proposed,and the general steps of extraction are given;(2)Based on the analysis of the existing fuzzy neural network structure,the network structure is optimized,the six-layer compensation fuzzy neural network structure of remote sensing image information extraction is proposed and the general model of remote sensing image extraction of compensation fuzzy neural network is established;(3)Based on the analysis of the training sample structure in the remote sensing image extraction of existing artificial neural networks,this paper proposes a method to generate training samples of high resolution remote sensing images provided by Google Earth directly from the segregation of the training sample set and the test set;(4)Based on the analysis of the existing single feature extraction method of remote sensing image,concluding the texture,color,shape and spectrum,a method of single feature extraction of remote sensing image information based on compensatory fuzzy neural network theory is proposed,and the corresponding compensated fuzzy neural network model is established;(5)Based on the analysis of multi-feature information extraction of existing remote sensing images,a method of remote sensing image information extraction based on compensation fuzzy neural network theory,such as dual feature combination,three-feature combination,and multi-feature fusion,is proposed,and the corresponding compensation fuzzy neural network calculation model is established;(6)Using MATLAB software as a platform,a single feature,dual feature,three-feature,four-feature fusion information extraction,image pre-processing and computational block program are designed to provide an effective calculation tool for solving remote sensing image information extraction;(7)Through the in-depth application research of the remote sensing information extraction method based on the compensation fuzzy neural network,including the single feature of the texture,color,shape and spectrum;and the dual feature combination of the texture and color,the texture and shape,the texture and spectrum,the color and shape,the color and spectrum and the color and spectrum;and the three-feature combination of the texture,color and shape,the texture,color and the spectrum,and the shape,color and spectrum;and four-feature fusion of the texture,color,shape and spectrum,the results showed that the effect of single feature extraction was poor,and the effect of 4 features fusion was the best;(8)The Object oriented information extraction method using the industry recognized software eCognition and the index method of NDTBI the author created is used to extract the remote sensing image information,and the comparative study was carried out;(9)The above research results have demonstrated that the information extraction method of remote sensing image based on compensated fuzzy neural network theory has been shown feasible.There is no need to do complex interpretation for remote sensing image in advance and no human intervention is necessary by the method,which has the superiority of short time consuming,fast calculation,high precision,automatic information extraction and convenient for practical application.The method provides an useful reference of an automatic,accurate and efficient information extraction of remote sensing image.