Change Detection Using IKONOS Imagery

Introduction

         Our environment is subject to many changes due to human activities and natural processes. Examples are the cultivation of nature, new infrastructure, urban development, movement of traffic, and changing coastlines, so in this assignment we focus target into change detection method that have had several way to perform this tasks such as Red-Green Method, Image Differencing, Image Ratioing, and Principle Component2. However, for monitoring purpose, remote sensing has advantage because of their capability to revisit the same area in several days. Satellite system especially, can give a non-expensive time series imagery, which uses the same sensor and spectral band consistently. IKONOS satellite system is quite new so there are still limited references found. The objective of this assignment is to explore the capabilities of IKONOS image in mapping and monitoring purpose. The study area is Pagatan in South Borneo.

          For the purpose of change detection; monitoring changes in earth surface’s objects and phenomena, multi temporal analysis: ie. Deforestation rate, and update spatial information database. However, before we going to do change detection we have to consider about the images have to be rectified, which that’s good and important idea that tried to make RMS error less than 1 pixel and involve, remove atmospheric effect, calibrated image for doing same DN between image date1 and image date2, etc.

Objective

1. Understand the main idea for performing change detection for the best result.

2. Learn to registration and rectification on the two images, and also understanding about why we have to calibration.

3. Learn to create a Virtual dataset of two rectified images.

4. Learn a simple change detection method displaying compatible bands of imagery of different dates in Red and Green color layer in RGB color mode.

5. Learn the band ratioing change detection method of dividing corresponding bands of imagery of different dates and display it as greyscale or pseudocolor image.

6. Learn to use the principle component for change detection method and also able to used formula for area calculation.

Data available

Metadata of these images can be read in the appendix

Background

         Change detection

         Change detection involves the use of multi temporal data sets to discriminate areas of the land cover changes between dates of imaging. The of change that might be of interest can range from short-term phenomena such as snow cover or flood water to the long-term phenomena such as urban fringe development or desertification. Ideally change detection procedures should involve data acquired by the same (or similar) sensor and be recorded using the same spatial resolution, viewing geometry, spectral band, and time of day. Often anniversary dates are used to minimize sun angle and seasonal differences. Accurate spatial registration of the various dates of imagery is also a requirement for effective change detection. Registration to within to pixel is generally required. Clearly, when misregistration is grater that one pixel, numerous errors will result when comparing the image.

         One way of discriminating changes between two data of imaging is to employ post classification comparison. In this approach, two dates are independently classified and registered. Then an algorithm can be employ to determine those pixels with a change in classification between dates. In addition, statistics (and change Map) can be complied to express the specific nature of the change between the dates of imagery. Obviously, the accuracy of such procedures depends upon the accuracy of each of the independent classifications used in the analysis. The errors present in each of the initial classifications are compounded in the change detection process.

         Another approach to change detection using spectral pattern recognition is simply the classification of multi temporal data sets. In this alternative, a single classification is performed on a combined data set for the two dates of interest. Supervised or Unsupervised classification is used to categorize the land cover classes in the combine image. The success of such efforts depends upon the extent to which “ change classes” are significantly different spectrally from the “no change” classes. Also, the dimensionality and complexity of the classification can be quite great, and if all bands from each date are used, there may be substantial redundancy.

          Temporal Image differencing is yet another common approach to change detection. In the image differencing procedure DNs from one data are simply subtracted from those of the other. The difference in areas of on change will be very small (Approaching zero), and areas of change will manifest larger negative or positive values. If 8-bit image are used, the possible range of values for the different image is-255 to255, so normally a constant (e.g. 255) is added to each difference image value for display purposes.

           Temporal image ratioing involves computing the ratio of the data from two image of imaging. Ratios for area of no change tend toward 1 and areas of change will have higher or lower ratio values. Again, the ratio data are normally scaled for display purpose one of the advantages to the ratioing technique is that it tends normalizes the data for changes in such extraneous factors as sun angle, shadow, etc.

          Whether image differencing of ratioing is employed, the analysis must fine a meaningful “ change –no change threshold” within the data. Compiling a histogram for the differenced or ratioed images data and noting that the change areas will reside within the tails of the distribution. A variance from the mean can then be chosen and tested empirically to determine if it represents a reasonable threshold. The threshold can also be varied interactively in most image analysis systems so the analysis can obtain immediate visual feedback on the suitability of given threshold.

         Change vector analysis is a change detection procedure that is a conceptual extension of image differencing; the basis for this approach in two dimensions. Two spectral variables (e.g. data from two bands, two vegetation components) are plotted at dates 1 and 2 for a given pixel. The vector connecting these two data sets describes both the magnitude and direction of spectral change between dates. A threshold on the magnitude cab be established as the basis for determining areas of change, and the direction of the spectral change vector often relates for the types of change.

IKONOS

            IKONOS have had 1-meter resolution and also had characteristic is orbit: 680 Km, sun-synchronous, 98.2 inclination, which it cans application to manage natural resource such as Water Resource, Land Resource, Environment& Hazards Management.

Theory and Methodology

Registration

            Every image has a Coordinate Space, which is defined by the Coordinate Space block in the image header file. The entries are: Datum, Projection, Registration Point coordinates, Image cell size, Coordinate Type Units (optional) and Rotation. In a RAW image no values have been assigned to Coordinate Space entries. Registering an image involves allocating appropriate values to these entries so that the image can be geographically positioned. See 'Coordinate spaces The Registration Point is a single cell in the image whose X, Y cell coordinates (usually 0, 0) are referenced to geographical coordinates or to those of the same point on another image. This is also known, as geo-referencing information Registration does not involve changing the image in any way.

Coordinate spaces

            The coordinate space of an image dataset defines where the image is located in the real world or relative to other images, which the coordinate space is specified by the following parameters in the header file: Map projection (Geodetic datum, Rotation.) and also it the coordinate space of an image is stored in the image header file.

Rectification

            Rectification involves changing the image to meet specific requirements. In addition to the Coordinate Space block, the image header file can also contain information on Ground Control Points (GCPs). These GCPs are points on the image whose position can be referenced to known coordinates or to the same points on a geocoded image. There are two different types of rectification are geometric correction and geographic transformation. The first one Geometric correction as described earlier in this chapter, digital image data can often contain errors in geometry due to the motion of the scanners, sensor characteristics, the curvature of the earth, or other factors. Geometric correction uses information such as GCPs and/or camera characteristics to rectify these errors. For the second one is Geographic transformation: the Geographic transformation involves changing the geographic position of the image by rotating it and/or transforming it to another map projection, datum or coordinate type.

            We can rectify an image to change its map projection or rotation. Data not in a specific map projection, for example raw data, can be geo-referenced to a known map projection or to another raw image by identifying control points. We have several type of rectification. However, in this assignment we have selected Polynomial (Control Point) - Image to Image that we can start tasks by selecting Geocoding Wizard from the Process menu. Note: Images in the same map projection in the same algorithm are automatically mosaic together.

Change detection requirement

        ท      Image registration and rectification to make both images be in the same map datum and projection. PO_53518.ers should be the master image (target image) since it has been rectified.

        ท     Image calibration. Check the mean DN values for the dark and bright features at the same location. Define polygons using raster region in drawing utility and then calculate statistics. Transform the second image linearly.

        ท     PCA transformation. For each image, make Principle Component #1 transformation to have image compression.

        ท     Make a virtual dataset that contains:

o         4 bands of PO_53518 data

o         4 bands of PO_59044 data

o         PC#1 of PO_53518 data

o         PC#1 of PO_59044 data

Change detection

• Red Green Difference Image
• Band Ratioing
• Principle Component # 2

 Previous          Next