| As a kind of complex probing technology, remote sensing can supply information ofnatural process and phenomenon quickly and effectively, thus reveal their dynamic changingpattern and forecast their developing tendency. By remote sensing, mass amount ofinformation and data can not only be acquired quickly, the scientific, accurate and promptanalysis of them can be reached. Remote sensing can not only supply information of somespecific area, but that of a whole area. It provides a kind of new and effective method ofmonitoring and mastering city wholly, multi-spectrally, multi-temporally and a scientificevidence and a technological support of city green land system planning.Information Extraction of remote sensing has five kinds.Classification means to recognize and classify objects from RS images by their spectral,special and temporal information.Change detection means to detect changes of objects from images of different time bytheir spectral information.Extraction of physical quantities means to get air component or temperature of objects byspectral information and get elevation by stereo image pairs.Index extraction means the process of calculating plant index.Identification of specific features means to recognize disaster condition, linear structureand historical remains of specific ground objects or status of land surface.With extended functions, it can realize GIS topological relation building, map connection,transformation between thematic map and vector data.This article explains city green land extraction from IKONO RS image by ERDASIMAGINE, to realize the automatic or half automatic extraction of city green landinformation, and to reach the statistical analysis of the city green land information integratinginformation of city population and city area, etc. To get a deeper understanding of computerrecognition of RS image through the classification process is the basic objective of thisarticle. It attaches importance to the effect on classification of two different bandcombinations from IKONOS RS 4 bands image. At the same time, comparison analysis ismade between the IKONOS image classification and the classification of TM image atdifferent time.The main contents of this research are as following.1,RS image preprocessingIn order to get more information that cat meet requirements, the preprocessing of RSimage data is necessary in RS application research, it is the prerequisite and basis to thedeeper research. In green land extraction, the preprocessing of RS image includes:A, the acquisition of basic information and background information of original RS imageand research areaB, geometric correction and projection transformation of RS imageC, sample area selection, acquisition and cutting of multi-spectral RS image2,RS image information extractionAfter the selection of two groups of band combination, color combination enhancementto the RS data was made. The target objects ware obviously shown from the backgroundimage. Two groups of band combination are red, green and blue band combination andinfrared, red, green band combination. In order to improve image resolution and betterclassification, we used an IKONOS panchromatic image of 1 meter resolution. Each image oftwo kinds of band combination is merged with the image of 1 meter resolution. As a result,the effect of mixed pixel is decreased.Next, the merged imaged were classified. Supervised classification was used. After testsof training area building, objects in images are classified in four categories, namely, groundsurface, building, water and green land. After the computer classification, lots of on-the-siteinvestigations are done to improve the classification accuracy.The two classifications indicate that the one that used infrared band is much better thanthe other one. Plants have better inflection in infrared band.Lastly, TM image data are used. Data of TM3, TM4 and TM5 band are combined together.Also, the supervised classification is used. The three classifications are compared, thedevelopment of green land use is analyzed and this is good to green land planning inShenyang city.At the same time, four conclusions are reached.First, the classification with infrared band image is better in green land informationextraction. The research to ground object inflection is good to the research of ground objectautomatic extraction and quantitative analysis.Second, errors of classification mainly come from the still existing phenomenon of thesame objects with different spectrum and different objects of the same spectrum. Due to thehigh resolution, the green land information is better shown and classified.Third, the classification accuracy is affected by the building of training area. The buildingof training area needs the knowledge of object inflection and experience. Otherwise, badbuilding of training area leads directly to the mass work of on-the-site investigations.Last, after the comparison of three classifications, classification with high resolution ismuch lively and detailed. And the comparison between classifications of images of differenttime, current and earlier, shows that the green land area of Shenyang is enlarged, especiallyin the areas along streets and inside living units and the integration of RS and GIS is animportant research field.The main goals that the research wants to reach are as following.First is the effect on classification accuracy of different band or band combinationselection. IKONOS has two types of RS data, panchromatic band image data andmulti-spectral image data. Resolution of the former is 1m. The band scope is from0.45~0.90um. Resolution of the latter is 4m.the multi-spectrum has four bands. The blueband spans from 0.45~0.53um. The green band spans from 0.52~0.61um um. The red bandspans from 0.64~0.72um. The near infrared band spans from 0.77~0.88um. The comparisonbetween two classifications from the two different band combinations is the main topic.Second is the statistical analysis integrated with data from other sources. With thedevelopment of RS technology, satellite image data with high resolution are now commonlyused in environmental monitoring. They have the advantages of accuracy, economy, and timeeffectiveness. This research also introduces the comparison between the classifications fromIKONOS image data with 1m-resolution in 2003 and TM image data with 20m-resolution in1999. It shows another kind of application and is another powerful evidence of RS imageapplication in green land information.At present, good classification of objects under shadow is not available. The small imagespeckle processing is also an obstacle to information extraction. Phenomenon of same objectswith different spectrum or different objects with same spectrum is a main factor to affect theinformation extraction accuracy. How can the image data with high resolution be better usedto change detection? These are the problems that need to be deeply discussed, which are notresolved or touched in this article. |
PDF Full Text Request