Research On Identification Of Urban Stock Constructive Land Based On Data Mining

Due to the defined connotation and types of urban stock constructive land are based on the public administration perspective in China, but not from the actual use of state of urban land, therefore, leading to the large area of recessive vacant and partially vacant land, which is not included in the urban stock constructive land. In addition, from the obtaining research information of urban stock constructive land, our scholars lack of enough attention and deep research. Although western scholars use GIS technology for the stock constructive land research analysis, they do not connect object-oriented remote sensing analysis, GIS technology with the data mining methods, such as neural network, fuzzy theory and so on. For that reason, this thesis will be based upon the urban actual land of state to define the urban stock constructive land. In addition, with the help of remote sensing image analysis, geographic information system technology, field exploration and professional theory and so on, data source from high resolution remote sensing image, cadastral database technology is to help for identification research.This thesis will demonstrate mainly from the following three aspects:(1) Using system analysis for the urban stock constructive land research at home and abroad, data mining and object-oriented method in the land science and relevant field’s applied research.(2) Start from the actual state of urban construction land, using ancestor land as basic unit to redefine the connotation of urban stock constructive land in our country, urban stock constructive land will be divided into vacant land, partially utilized land and underutilized land and set up the standard identification respectively.(3) Shaoxing City, Zhejiang Province is selected as the research area. According to the characteristic of remote sensing data and database data, reference identification standard, it will put forward identification method and recognition process to the urban stock constructive land, using an object-oriented method based on image segmentation, data mining methods based on decision tree, supported vector machine, fuzzy theory, BP neural network and spatial analysis methods, and conducts positive analysis.In this thesis, the results indicate:1) From the actual state of urban constructive land, using ancestor land as basic evaluation unit to redefine the connotation and divide types of urban stock constructive land has theoretical and practical significance. It not only considers the apparent state of land utilization, but also considers the embodied degree of function value of land attachments. In addition, it not only fits the connotation of land value, but also reflects the up-to-date state of urban stock constructive land information.2) On the condition of the object-oriented vacant identification, fully uses objects’ spatial features, spectral features and textural features and applies the supported vector machine and fuzzy based on the rules classification model to conduct quasi vacant land identification, overcoming "Salt and Pepper effect" and then increases the classification accuracy of image. The overall classification accuracy and Kappa coefficient of supported vector machine classification model are 87.76% and 0.86. Moreover, the overall classification accuracy and Kappa coefficient of fuzzy based on the rules classification model are 91.49% and 0.90.3) On the condition of the identification of vacant land and partially utilized land which are based on the spatial analysis, this thesis brings in the idleness rate of ancestor land concept, applying the spatial analysis based on the principle of overlay analysis to carry out identification for vacant land and partially utilized land in the research area. From the identification results perspective, vacant land and partially utilized land’s number and proportion accords with the basic line of the research area.4) On the condition of the identification of underutilized land which is based on the neural network, the error of fitting of the learning samples in the BP neural network model is 9.7×10-6, and the classification accuracy and Kappa coefficient of the testing samples are 100% and 1 respectively. When we use BP neural network model to identify the predicted data, identification results and image or field investigation results are in full accords, which accuracy rate is 100%. This research results show that BP neural network model overcomes biggish shortcomings of indexes subjectivity of the weight decision in the Comprehensive Evaluation Method and Delphimethod, and can be more objective to identify underutilized land and full developed land in the urban stock constructive land, which is a relatively objective method. From the identification results perspective, underutilized land and full developed land accords with the basic line of the research area.