Change Detection Of Construction Land From Low-Altitude Aerial Imagery

The thesis focuses on researching how to detect changes of construction land, especially the increased construction land, from low-altitude aerial imagery. A typical region of Jintan City, Jiangsu Province is selected as the study area. The proposed method can be applied to improve the efficiency of dynamic monitoring land resource.Construction land is one of the core elements of industrialization and urbanization. So it is important to grasp the changes of construction land to improve the utilization of land resources and coordinate contradictions between cultivated land protection and economic development. Low-altitude remote sensing platform (UAV, unmanned airship) is characterized as tactical, real-time, and low cost. Compared with the satellite remote sensing, the low altitude remote sensing platform breaks the limits of cloud coverage and satellite visit circle. Detecting changes of construction land from low-altitude aerial imagery is now a new way of dynamic monitoring of land resources.A new method for change detection of construction land was proposed, which integrated the pixel analysis and object recognition in a unified framework. First, according to the characteristics of low-altitude aerial imagery, the spectral information and texture information were combined to design a integrated difference method that getting land use change information at pixel-level. Then the mathematical morphology method was used to improve the accuracy and the integrity of change detection results. Moreover, it helped to achieve the transition from pixel-level detection to the object-level recognition. Finally, the change detection result were further validated by the means of texture analysis at object level, and the changed objects of construction land were recognized from all the changed objects.The main contents and results of the research are presented as follows:(1) Low-altitude aerial imagery contains more structure and texture information because of its high spatial resolution. However, its spectral resolution is relatively low as there are only three spectral bands. Hence, the change detection method should take full advantage of texture information of the low-altitude aerial Imagery. So the appropriate texture should be selected on the basis of spectral features. Then an adaptive threshold segmentation method is used to get the change information from spectral and texture bands, respectively. Finally, the change detection results of each band are combined to represent all the changes. The experimental results show that texture information can reflect the detail information within changed regions. So it can effectively reduce the fragmentation of the change detection results while improving the detection accuracy. The integrated adaptive difference method proposed in this research can get a result with a higher degree of accuracy and lower rate of undetected when compared to other commonly used change detection methods that based on direct comparison.(2) Affected by the registration error and the internal differences in objects, the change detection results were more fragmentary. In such case, on one hand the change information is not expressed completely, on the other hand many meaningless and small changed areas were generated. So four steps, including morphological close, holes filling, morphological open, and small region removing, were taken to optimize the results of change detection to remove the dummy changes and make the change areas much more complete. It also facilitates to get the changed objects and achieve the transition from pixel-level detection to the object-level recognition. (3) Due to the local subtle structural differences, there are still some of pseudo changes at the pixel-level change detection results. So it needs to be further filtered out at the object-level. Firstly, the changed objects are generated by tracking the connected changed pixels. Then, the pseudo changed objects are filtered out according to the texture difference of objects. After validation, there are still different types of change information in the changed objects, not only the construction land changing objects, but also non-construction land changing objects. The proper texture features are selected for homogeneous and heterogeneous construction land objects respectively to recognize the construction land objects from all the changed objects.As a whole, the comprehensive application of texture and spectral information and the optimization of the change detection results can effectively reduce the undetected rate. The filtering of changed objects and the recognition of land construction objects can improve the accuracy of change detection of construction land. Experiments prove that the proposed method can get the geometric and physical attributes of the changes of construction land quickly, help to improve the efficiency of dynamic monitoring of construction land.