Study On The Classification Of Remote Sensing Image In Geographic National Condition Monitoring

The key technology of information extraction of surface covering the geographicalcondition monitoring in the project, is how to quickly and accurately obtain objectinformation of remote sensing image contains. This article uses the eCognition software forautomatic image interpretation,the process of interpreting is divided into two steps: the firststep is to select the optimal combination of parameters for multiscale segmentation, the resultsof segmentation directly affect the quality of the final the results of interpreting. In order toobtain the best segmentation combination of parameters, a detailed study on how to select theparameters of its division and experiments in this paper. The second step is to generate animage segmentation result according to the image object layer object properties andclassification rules set by writing.Since the automatic interpretation of the results of the data has serious jaggies, and theaccuracy can not meet the needs of land cover projects. Hence we need to manually modify.Based the aliasing problems, this paper modeled by using ArcGIS Model Builder modelgenerator to resolve. After manually through the "stream editor"(Streaming) model ismodified, there is a common error that adjacent surface has the same properties, manualinspection is very difficult to find the problem.it’s sovled through the scripting languagePython writting in a special inspection tool. Finally, by the counting the area of interpretingthe results and modifying the results of all artificial feature type, we can interpret the correctlyfeature type area, then turning on the automatic interpretation evaluate the accuracy of theresults.The researth in this paper is based on biased northern Zhiduo, Qinghai Province, whichuses2.4meter spatial resolution of QuickBird remote sensing images, including four bands:red, green,blue, NIR. By Combining with object-oriented classification software eCognition8.8and spatial analysis software ArcGIS9.3to the geographic national condition monitoringof land cover information extraction production practice. The reseach we get the Overall accuracy is up to93.90%with strong applicability.The innovation lies in this article, to some extent, it’s realizes the human-machineinteractive vectorization, Greatly reducing the manual workload and improve the efficiency ofdaily production. Research findings can provide a very good reference value to the lategeography situation monitoring land cover acquisition.