Object-oriented Information Extraction Based On High-resolution Remote Sensing Images

In recent years, the rapid development of new sensor makes more and morehigh-resolution remote sensing images, and brought a richer remote sensing imageinformation. How to take advantage of high-resolution remote sensing images richerstructural information, as well as related information domain object texture information,high-resolution remote sensing images automatically and accurately, thematic elements ofinformation extraction efficiency has become an urgent need in the field of remote sensingtechnology problems.Spectral information using traditional methods such as information extraction element,which is the basic unit for processing pixel, more suitable for low-resolution remote sensingimages, and for a strong spatial information, texture information between domain objects andassociated information the high-resolution remote sensing images based on traditionalmethods like element to its good classification can’t be achieved, but also lead to more serious"salt and pepper phenomenon", and "with the spectrum of foreign body" phenomenon and the"same thing different spectrum" phenomenon in the information extraction results are veryobvious. The use of object-oriented image information extraction methods to improve theaccuracy of high-resolution remote sensing image information extraction and efficiency in thefield of remote sensing has become a hot topic.In this paper, Shang Xi Chang An District, West Village, pomegranate as the study area,with WorldView-2satellite image of the experimental data. Completed using object-orientedmethod for high-resolution remote sensing images to extract information thematic elements,and pixel-based and object-oriented method of information extraction accuracy to make anobjective evaluation. The main contents are as follows:(1)Multi-spectral remote sensing image fusion method panchromatic band and thecorresponding study conducted to evaluate the HPF integration from both subjective andobjective aspects, IHS fusion, wavelet transform fusion and pan sharpening fusion, fusion research proves pan sharpening the method can better maintain the spectral properties ofremote sensing images and effectively enhance the texture and detail information of theoriginal multispectral images, thematic elements of information extraction has laid a goodfoundation.(2)By previous studies and the experimental data multiscale segmentation experiment,this paper summarizes the method to evaluate the optimal parameters of multi-scalesegmentation, information has been extracted from different thematic elements segmentationparameters, and the use of multi-scale edge detection algorithm to improve split it caneffectively improve the accuracy and effectiveness of multi-scale segmentation. Thematicelements of information extraction provides a reliable segmentation accuracy, improvedthematic elements of the overall extraction accuracy.(3)Multi-scale experimental data to improve segmentation derived thematic elements ofinformation extracted optimal segmentation parameters. In dividing the scale parameter is150, the spectral factor of0.2, firmness factor of0.5for the optimal parameters to extractwater. In dividing the scale parameter of100spectral factor of0.2, firmness factor extractionbare land, roads, woodland, arable land is the optimal parameter of0.5. In dividing the scaleparameter is50, the spectral factor of0.1, firmness factor of0.6for the extraction of buildingoptimal parameters established by experimental studies thematic elements for different levelsof information extraction and classification hierarchy.(4)Spectral information utilization of images, spatial information, texture informationand related information in the field of information extraction target different thematicelements. Through thematic elements characteristic description and combination, theestablishment of spectral information image object to quantify the expression model of spatialinformation, texture information, and related information in areas such as object of eachfeature to achieve a structured thematic elements of characterization, and visualization,established the image information extraction rule sets, rule sets optimized to achieve a rapidand accurate for the different thematic elements effective information extraction.(5)Based on the results of pixel and object-oriented methods to extract informationobtained by the use of the confusion matrix to calculate the accuracy of informationextraction, to get the overall accuracy of object-oriented is94%, kappa coefficient of0.89,based on the overall fine pixel is73%. kappa coefficient of0.70, an objective analysis derived object-oriented approach is more suitable for high-resolution remote sensing imageinformation extraction.