Granularity Analysis And Urban Building Change Detection

Remote sensing change detection refers to the identification of an object’s or phenomenon’s position change at the different time, building change detection includes three types:new, demolition and reconstruction. The significance of building detection is that updating the urban geographical spatial database, which reflects the rationality of land use, indicating the direction of the regional planning, In a word, it plays an important role in urban informatization management. The first step is to confirm the size while mining spatial entity in GIS. (that is, the basic unit when organize data), construction change detection based on pixel (including pixel neighborhood) and object granularity. For detecting particle size in pixels, use pixels of the spectrum, such as texture, neighborhood statistics information; Based on object size, first of all need access to remote sensing image segmentation, figure and boundary can use the position of the object, area, adjacent degree etc. Characteristics.Therefore, the research content mainly includes three parts:(1) the design and test used in the classification of urban underlying surface descriptors, main purpose is to extract maximum building objects.(2) the design of architectural change detection method based on particle size analysis, including adjacent degree analysis method based on object, change the layer weighted joint analysis method (pixel).(3) to remove pseudo changes, including pseudo change remote sensing image when shooting building tilt, caused by the change of the nadir point image scale caused by different construction outline do not match and shadow noise.Through the area of filtration, adjacent degree principle and multidimensional feature space weighted denoising method will remove these pseudo change.The article introduced and compared two kinds of object detection method based on adjacent buildings change analysis (object) and multi-dimensional change feature space weighted joint analysis method (pixel).The precision is low, the average accuracy is 37.6%, and virtual screening rate is too high (average detection rate of 34.2%), mainly from the construction, figure is incomplete and the error because of proximity analysis method based on object requires a complete object boundary.And if starting from the differences of two realities image characteristics, design of multiple image information as the change of multidimensional space change characteristics, obviously improve the efficiency, the average accuracy can reach 87.7%, this method is called the dimensional change feature space weighted joint analysis.There are five conclusions:(1) using the spectral characteristic of urban underlying surface composite classification accuracy than using separate spectral characteristic of precision, the spectral characteristics of composite is composed of multiple independent spectral characteristics (such as:R, G, B, H, S, V) to form the compound descriptors (such as NDVI, NDSV);(2) the combination of spectrum and texture feature of urban underlying surface classification effect is superior to single use texture or spectral characteristics;(3) for the classification of urban underlying surface, using a priori knowledge of supervised classification more efficient than unsupervised classification;(4) adjacent degree analysis method is used to filter caused by nadir point change of pseudo change effect is good.(5) the use of multi-dimensional change feature space weighted joint analysis accuracy than based on object adjacent degree analysis precision is higher about 50% on average.