Abstract


Man Made Change Detection Using IRS Images


Abolfazl Aghaee Meybodi
Tehran University,
Iran
Email: ab_aghaee@yahoo.com


Dr.Mohammad Reza Saradjian
Supervisor
Tehran University
Email: sarajian@ut.ac.com


Monitoring of urban expansion due to human activities using remote sensing data is necessary in natural resource management. Land change monitoring because of man-made constructions is also crucial in military remote sensing applications. By definition, change may be attributed to any difference in corresponding pixel values within multi-temporal images. However, the change detection methods used and their sensitivity to detect the change may have an impact on the results. Also, the images used regarding their spatial and spectral characteristics may affect the accuracy of results.This research is involved with change detection of land particularly man-made objects using IRS images within a specified time interval. It is first required to detect the components such as buildings as man-made objects and then perform the change detection accordingly. In order to increase the efficiency of the process in the case of mass data analysis, the automation of the process has been taken into consideration. Therefore, unsupervised classification algorithms have been chosen for separation of buildings from background pixels automatically.In order to enable the IRS images to have the potential of showing individual buildings, IRS-LISSIII bands have been fused with IRS-PAN. Then, a normalization process has been applied to avoid false change detection due to differences in imaging conditions. In the next step, in order to simplify the images, vegetation covers and water bodies have been detected and masked out. The remaining pixels of the image have been regarded as either background or non-background objects. The non-background objects have been assumed as man-made objects. Because of the closeness of the buildings' pixel values to the background pixel values, their separation produced undesired results.Using discriminant function and performing linear transformation on images, the contrast has been increased resulting in higher accuracy classification. In this regard, weight coefficients have been calculated by discriminant function and applied on the images which contain no vegetation and no water pixels. As a result of applying the weight coefficients, the contrast between man-made and background pixels has been increased. After performing unsupervised classification algorithm, the man-made objects have been detected in temporal images with relatively high accuracy. Finally, the change detection process has been applied based on differencing of classification map of temporal images. After eliminating edge of features falsely detected as change due to the some inaccuracies in the geometric registration and fusion of images, the overall process presented convincing results in the man-made objects' change detection within temporal images.