Thresholding Algorithm Optimization for Change Detection to Satellite Imagery

To detect changes in satellite imagery, a supervised change detection technique was applied to Landsat images from an area in the south of México. At first, the linear regression (LR) method using the first principal component (1-PC) data, the Chi-square transformation (CST) method using first three principal component (PC-3), and tasseled cap (TC) images were applied to obtain the continuous images of change. Then, the threshold was defined by statistical parameters, and histogram secant techniques to categorize as change or unchanged the pixels. A threshold optimization iterative algorithm is proposed, based on the ground truth data and assessing the accuracy of a range of threshold values through the corresponding Kappa coefficient of concordance. Finally, to evaluate the change detection accuracy of conventional methods and the threshold optimization algorithm, 90 polygons (15,543 pixels) were sampled, categorized as real change/unchanged zones, and defined as ground truth, from the interpretation of color aerial photo slides aided by the land cover maps to obtain the omission/commission errors and the Kappa coefficient of agreement. The results show that the threshold optimization is a suitable approach that can be applied for change detection analysis