Before doing this lab, please read the Remote Sensing Core Curriculum Change Detection Module.

I. Introduction

In this lab we will do a change detection using the Landsat data for the Raleigh area from 1988 and 1994.

We will attempt to determine changed areas using a simple differencing of band four. Taking the difference of band four is a way to explore areas that have different near infrared reflectance for the two dates.

In addition to discussing change detection, we will also use this lab to introduce the "Spatial Modeler" in Imagine.

II. Pre-processing Considerations

As described in Jensen's text, pre-processing is a critical step in any digital change detection. For the purposes of this lab, however, we did not do any geometric rectification. Similarly, the data are left in Digital Numbers (DN) and there was NO radiometric nor atmospheric adjustment. The pre-processing steps were intentionally not done to demonstrate their importance. The result is that areas which have NOT changed may NOT have a difference value of zero. This should be kept in mind when interpreting the difference image.

III. Image Differencing

As mentioned in the introduction, to do the image differencing we will use Imagine's "Spatial Modeler." The spatial modeler is a graphical programming interface for Imagine. It allows you to manipulate an image or set of images with various functions and routines. The example below provides an introduction, but note that there are many other functions and options within the spatial modeler.

To get into the spatial modeler, select the Modeler button from Imagine's main menu bar. This will give you the option of choosing MODEL MAKER... or Model Librarian... . Since we are interested in making a model, select the MODEL MAKER . This will give you the "Model Maker" window as well as the Model "tools" window:

As you move over the different buttons in the toolbox, a description of that tool will appear at the bottom of the Model Maker window. Find and click on the tool that will Place a raster object in the model . Then click anywhere in the Model Maker window. Once you have done this you will see an icon similar to the tool button in the Model window. Double Click on this icon and you will be able to select an image for input within the following dialog box:

In this dialog box choose the tm_data/raleigh88_alb.img image file, accept the default parameters, and click OK . Repeat this process to add tm_data/raleigh94_alb.img as another raster input for the model.

Now, lets do something with these two input images. Click on the tool that will Place a function in the model and click anywhere in the Model Maker window. A circle should appear..

Next we will link this function to the input images. Select the tool that Connect inputs to functions or functions to outputs and, starting on one of the images click and hold on an imput image and slide over to the "function" icon. Do this twice so that each image is attached to the one function.

Note, with the pointer tool selected you can click and hold on any of the icons in the spatial modeler to move them around. The link should remain.

Now, we will define the function.

Double-click on the function icon to get the functions dialog box:

For the following steps, don't be concerned if the $n_ numbers are different from what are in your model. Just concentrate on selecting the correct images and bands. In this box, first click on $n1_raleigh_94(4) from the "Available Inputs:" window, then hit the "-" button from the calculator functions, and then click on $n2_raleigh_88(4) . Note that you could also type this function in the function box by using the keyboard. This will cause band 4 from the 1988 image to be subtracted from band 4 of the 1994 image. Click OK to close the function box.

Now, lets put the output of the function somewhere. Again click on the tool that will Place a raster object in the model , click the raster object into the model and then attach the function to this new raster object with the connection tool. Double click on the raster object so to define the output. Within the dialog box check the Delete if Exists box, change the data type to Signed 8-bit data, and make sure that the Ignore 0.0 in Stats Calculation is NOT turned one. (Can you guess why?) Give your output difference image a name and hit OK . This completes the steps needed to construct a model for a "band-four differencing".

When you're finished your model should look like this:

To run the model, click on the Red lightning bolt.

IV. Finding Changed Areas: classifying the the difference image

Now that we have a difference image, we need to use this image to determine which areas have changed. First, open the difference image in a viewer and look at the image statistics using Utility / Layer Info. What do the mean, min, max, and standard deviation represent in this image?

We can further analyze the difference image by doing an unsupervised classification. Specify ten or more classes. Because the unsupervised classification will be working with only one band, the extreme classes will represent the extreme values for that band.

Review Questions

1. What do you think the extreme high values in band four differences represent? How would you interpret extreme low values?

2. In the original difference image (i.e. unclassified), why are there white edges around the bodies of water?

3. Go back to the spatial modeler and change the function to do another form of change detection. Run the model and explain what you did and how you interpret the results.

4. In general, what are the effects of not doing any atmospheric corrections (see Jensen)?

5. Describe at least one other change detection technique.

6. Using the "View" button described in class, examine the underlying model for a function that interests you. Modify the model to do something that you think is cool. Describe what you did. (Note: Your focus here should be on learning the modeler. You don't necessarily have to do anything that has a real application. Have fun.)

Go back to the FOR753 Home Page