A Study Of Expert System For Linear Texture Understanding Of Remote Sensing Image

There are many problems in remote sensing imagery automatic interpretation needto solve by following approaches:Firstly, the existing methods of interpretation depend on imagery spectral charactersexceedingly but not mine spatial characters and other attribute characters sufficiently.Secondly, improve extracting precision according fusion utilized of multi-factor,multi-sourced remote sensing data and multi-sourced information.Thirdly, divide knowledge into different levels and fusion them according toknowledge level. Then, realize advanced understand to remote sensing image basedmethods of manual neural network and expert system.Finally, study mechanism of remote sensing from land surface to imaging process inorder to offer theory foundation for remote sensing imagery pattern recognitiontechnique.The paper suggested that it is an efficient approach for solving some existentproblems in automatic interpretation to system of remote sensing image understanding ofexpert system. At present, it conquered the disadvantage of low precision for pixelstatistic classification method and math classification. The disadvantage represent thatthose methods takes pixel as unit, which hardly make use macroscopically knowledge toprocessing of microcosmic unit. These methods have poor precision to classify because itonly applies simple spectral statistics and math knowledge but not the relation ofcontiguous spectrum characters (the relation form hue, color, texture, size, locality andcorrelation, etc.). It hardly solves the indefinite problems such as the same spectrumcorrespond to different object and the same object correspond to different spectrum onlyapply spectral characters for image. In the opposite way, the expert system of remotesensing image for linear texture realized a simulation of manual interpretation based oninterpreting expert according to make the best of feature object for image including thecompounding of spectral feature object, spatial feature object (feature class of geometryand analysissitus), texture feature class and attributing feature class (class knowledge forland surface objects).Remote sensing understanding is an efficacious approach to realize intellectualizedunderstanding and roboticized interpretation for remote sensing image. Image is incategory of spatial technique. Understanding is an important field of artificial intelligence.The integration of image and expert (understanding) is an efficacious combine fortechnique of space and artificial intelligence. That is to say, remote sensingunderstanding is to seek answers to problem of spatial field by applying artificialintelligence. The expert system considered as an embranchment of artificialintelligence. This paper developed an expert system of remote sensing imageunderstanding for linear texture under the background. It mainly used to study onautomatic classification extraction for linear texture on Landsat TM image.At the develop course to expert system of remote sensing image understanding forlinear texture, the paper did requirements analysis by means of software engineering andknowledge engineering, organized and arranged the whole system based present idea ofproject management, and interpreted remote sensing image interpretation which containsimple interpretation (simple analysis to image characters) and deep-leveledinterpretation (integrative application to remote sensing geosciences analysis method)made by expert, while sufficiently considered characteristic of remote sensing imageinterpretation such that the remote sensing image which is used is the planar informationexpression for reality and its spatial status of the objects.Knowledge engineer could obtain some knowledge used to interpret from book,literature and interpret expert of remote sensing image, but those knowledge is notadequate to the whole perplexing interpretation process. The paper acquired interface bynatural language based rules and humanness knowledge used to interpret, which solvedsome problems of interpretation expert knowledge acquiring. For the style of knowledgeacquire, repository automatically transferred the rules to the definition of object-orientedclass structure and message according to questionnaire investigation (includemultiple-choice questions, simple numerical value questions and questions of filling) ofnatural language based on rules. At the designing of uncertain reasoning, on the one hand,with the dependability theory we use the result of indefinite nature calculation to presenterror transfer and accumulation in the reasoning. On the other hand, because the interpretprocedure is a course of hypothesis and certification, here we used bi-directional messagereasoning manner, that is reverse reason selective object from the normal selective objectunder the drive of message. Besides, we also developed a system of feature extraction inremote sensing image that has the function of feeding blood (fact and evidence) forexpert system of remote sensing image understanding. In the system, we searched andmade use of a series of relative moderate pattern recognition algorithm and formedalgorithm mode. Expert system can acquire fact and certification for image according tothe feedback information transfer system of feature extraction in image momentarily atthe process of reasoning. For this, we put forward to expert system, image processingsystem, feature extraction system, integration of GIS data and model manner. We adoptedtwo integration manners of model and data at the course of those systems integration. Weselected data integration manner in the integration of expert system of remote sensingimage understanding, image processing system, feature extraction system and GIS, andmodel integration manner in the integration of expert system and feature extractionsystem, also feature extraction and GIS. Model integration manner realized spaceanalysis in feature extraction system and feature collection in expert system.The paper made large numbers of experiment use the experiment system wedeveloped. The results suggested that the precision is acceptable. The system canautomatic create experiment report including existent problem in the experiment. Thereport will be supplied to user as a text file. Besides, it should be a scale standard forquality and capability of expert system of remote sensing image understanding to theutilize level (veracity and rationality) of knowledge.The realization of automatic interpretation need suffer an endless course becauseautomatic interpretation for remote sensing image is complicated system engineering.Moreover, the study on expert system of remote sensing image understanding is in aperiod of unceasingly explore and development.