|
Urban greening is an important part of the urban ecosystem, with high spatial resolution satellite data is timely, accurate access to urban vegetation (urban vegetation type, distribution and structure), can provide the basis for quantitative analysis and evaluation of ecological benefits for the city, to meet the city Green construction and management of the department's needs. IKONOS images, explore select optimal segmentation scale based on the object-oriented method for urban vegetation classification, to build a class hierarchy of urban vegetation extraction of urban vegetation classification; accuracy is estimated to increase the amount of urban green BP neural network green biomass estimation model based on genetic algorithms to optimize the model, and reference factor to improve, this paper studies the content and conclusions are as follows: 1 image preprocessing for identifying urban vegetation in smaller type, improve vegetation classification accuracy, image fusion to enhance image interpretation. Analysis and comparison of the principal component transform 3,4-band fusion effect to best extract can be used for urban vegetation; building shadows and terrain affect the high-resolution satellite imagery, seriously interfere with the spectral information of the image Feature this article to take a different building shadow and hillshade were correct: the shadow of the city buildings, the use of object-oriented method of extraction, image segmentation Lambertian model correction; combined DEM Lambert model of Purple Mountain Hillshade, statistical models and experience were corrected, the results show that the the Lambertian model existed correction phenomenon, empirical statistical model calibration can achieve the desired results. 2, the use of object-oriented methods for automatic classification of urban vegetation, the experimental method to determine the optimal segmentation of urban vegetation scale; build urban vegetation class hierarchy based on commercial software, application object's spectrum, texture, and context information to achieve classification of urban vegetation, the overall classification accuracy of 85.5%, Kappa coefficient was 0.826, achieved better classification results. Compared to the commonly used classification based on pixel classification, object-oriented classification method can achieve higher accuracy. 3, the use of remote sensing information as urban vegetation green biomass data source, this paper, based on neural network estimation model, the estimation parameters change, select vegetation index as well as environmental factors, the elevation factor as independent variables; using genetic algorithms weight threshold of BP network optimization, genetic neural network optimization estimation model. The experimental results show that the added environmental factors, the elevation factor can improve the accuracy of estimation of green biomass; genetic algorithm to optimize BP neural network weights, thresholds, enabling network convergence to a global optimal solution, and to improve the stability of the network training.
|