Applying the chi-square transformation and automatic secant thresholding to Landsat imagery as unsupervised change detection methods

Abstract. In the performance of change detection analysis, the change/unchanged pixel categorization is usually made through empirical methods or trial-and-error manual procedures whose reliability may affect the results. To detect land-cover changes, an unsupervised change detection technique was applied to Landsat images from an area in the south of México. At first, normalized surface reflectance, principal components, and tasselled cap images were used to develop the chi-square transformation (CST) applied to each kind of image organized in absolute and relative values and thus, obtain the continuous image of change. Later, the histogram secant technique was applied to change images to automatically define the thresholds and categorize as change/unchanged the pixels. Finally, to assess the change detection accuracy, 86 polygons (14,512 pixels) were sampled, classified as real change/unchanged sites, and defined as ground-truth, from the interpretation of color aerial photo slides aided by the land-cover maps to obtain omission/commission errors and kappa coefficient of agreement. The results show that the CST and automatic histogram secant thresholding are suitable techniques that can be applied for unsupervised analysis change detection.