City Building Recognition Based On High-resolution Remote Sensing Image

The method based on remote sensing image processing and feature analysis has been one of the main access to object information, which has been widely used and promoted. Since the twentieth century, the rapid development of platform and increase of remote sensor’s diversity and accuracy make resolution better in space, spectrum, radiation and time, also object information be more exquisite and rich. Which in return also improve the accuracy of ground object’s recognition and feature extraction. Buildings and roads account for about eighty percent of ground objects in city high resolution images, so recognition and extraction of buildings account for a large proportion. City buildings play an important role in city planning, operation, and management, and recognition of city buildings is one of the key technologies of digital city construction in the future. So the research of building recognition has a certain significance to the urban development.The paper firstly introduces SQL Server 2008, includes relational model characteristics, relationship standardization, database design principles. Secondly, experimental area of remote sensing image was chosen and descripted, object characteristics are analyzed based on spectrum, geometric, textture, related characteristic parameters are extracted, table structure are designed, table relationship are related, data is storaged, datebase of building sample is established. Then supervised classification algorithms which include minimum distance classifier, bayesian classifier, neural network classifier and the integrated classifier of different weights based on multiple classifier fusion are used in classification of buildings on the basis of database of ground object spectral characteristics, and accuracys of all classification results are evaluated and compared by confusion matrix. Lastly, postclassification image with the best accuracy is choosed, small spots are eliminated, buildings are extracted by means of binary image, small connected regions are eliminated, edge position information of buildings are extracted and marked in the original image.Experiment has proved that datebase of building sample can storage and manage characteristic parameter, also has a good adaptability of city building diversity, what’s more, it can provides reliable basic data used in building recognition and spatial analysis for its own image. The integrated classifier of different weights based on multiple classifier fusion assign weight according to the accuracy of each classification, has a strong a daptability, which meet human cognitive habits for knowledge.Datebase of building sample and the integrated classifier of different weights based on multiple classifier fusion have a good effect in city building recognition, which is significant for research to some extent.