Remote Sensing: Lab 4: Miscellaneous Image Functions

Lab 4: Miscellaneous Image Functions

Goal and Background

This exercise was primarily designed to allow us to become more familiar with miscellaneous image functions within ERDAS Imagine 2015. There were seven main objectives involved with this lab. These objectives involved applying our pre-existing knowledge to hands on problem solving with optimizing spacial resolution, characterizing a study area from large satellite images, radiometric enhancement techniques, linking satellite imagery to Google Earth, re-sampling satellite imagery, image mosaicking, and binary change detection with graphical modeling.

Methods

Part 1 : Image Subsetting of Study Area (Creation of area of interest AOI)

There are two methods for creating a study area. The first method is creating a scene/study area through the use of the Inquire box in ERDAS Imagine. The second method is to delineate an area of interest (AOI) and use that to subset and chip. When using the first method, I added the inquire box and from there used the subset and chip tool to create my final study area defined by the area of the inquire box (Figure 1).

(Figure 1)

The second method used a shapefile to create AOI. I first imported the shapefile which contained two Eau Claire counties. I selected the two counties and used "paste from selected object" which created a dotted line around the selected counties. I then saved this boundary as an AOI file and used subset and chip with the saved AOI file to create my final area of interest using the second method of image subsetting (Figure 2).

(Figure 2)

Part 2 : Image Fusion

This part of the lab focused on optimizing spatial resolution for visual interpretation. Optimizing spatial resolution involved using two images with the Pan Sharpen tool. With the two images in separate viewers, I used resolution merge under pan sharpen. I then put my desired high resolution input and multispectral input files and specified the output. This resolution merge used a multiplicative method and a nearest neighbor sampling technique. After the resolution merge was complete, it was clear that the pan-sharpened imaged had an optimized resolution and was much easier to visually interpret than the pan sharpened image compared to the input reflective image. If zoomed in to the pixels it is clear that the newly pan sharpened image had much smaller pixels than the original input reflective image.

Part 3 : Simple Radiometric Enhancement Techniques

This section of the exercise familiarized us with radiometric quality and radiometric enhancement techniques. I used an image of Eau Claire with poor radiometric quality (Haze) and used the Haze Reduction tool under "Radiometric" in ERDAS Imagine to improve the overall radiometric quality of the image and reduce haze. The new haze reflection image no longer had a thin layer of haze over it and was much clearer than the input reflective image. The new image seemed to be much sharper and had a stronger shade of red than the pink color of the input reflective image. Shapes were much easier to see as well due to the haze reduction.

Part 4 : Linking Image Viewer to Google Earth

ERDAS Imagine allows the linking of images from their software to Google Earth. This has many benefits including serving as an image key for image interpretation. Google Earth serves as an excellent selective image interpretation key. It is a selective interpretation key because it has supporting text and an image analyst could match features with it. The process of linking or syncing images with Google Earth is not terribly complicated and uses the "Connect to Google Earth" tool. I brought an image into ERDAS Imagine and used "Link GE to View" and "Sync GE to View" to link/sync my image to Google Earth Pro.

Part 5 : Resampling

Resampling images involves the process of modifying the size of the image's pixels. I used two methods: nearest neighbor and bilinear interpolation. I used an image of Eau Claire and the "resample pixel size" option under Spatial tools . With the nearest neighbor method, there seemed to be no difference between the original image and the resampled image The pixels appear to be the same size and the color of the new image is the same as the input image as well. However, the pixels in the output image using bilinear interpolation were smaller than the original input image. The smaller pixels made the image look softer and less pixelated when zoomed in with the views synced between the input image and the output image.

Part 6 : Image Mosaicking

Image mosaicking is necessary when a study area exceeds the spatial extent of one satellite image. The first step in image mosaicking is to put both images adjacent to each other in one viewer. I used two methods for mosaicking, which were using Mosaic Express and then using Mosaic Pro. Mosaic Express involved inputting the two images and generating an output (Figure 3). Mosaic Pro had a more advanced routine and was not as quick of a process. Mosaic Pro is a little more complicated in which you are able to adjust radiometric properties at the area of interesection of the two images. Adjusting these properties is important so that there will be a smooth color transition from one image to the next (Figure 4). Comparing Figure 3 to Figure 4 we can see how much of a difference adjusting radiometric properties can make. Figure 4 has a smooth color transition, while figure 3 does not.

(Figure 3)

(Figure 4)

Part 7 : Binary Change Detection (Image Differencing)

This section of the lab involved using the brightness of pixels to detect binary changes in an image throughout a given time interval (1991 -2011). The first step is to define a change-no change threshold. I did this by using "two input operators" under "two image functions" under "functions". This does not show change however due to the fact that the change threshold needs to be defined by the user. To do this, we opened the Image Metadata interface and clicked on Histogram. We used a rule of thumb to determine threshold on the histogram ( Mean + 1.5 standard deviation). I put the cursor in the middle of the histogram and took note of the value. I then used the rule of thumb given and found my upper point for change-no change threshold. The lower point for the threshold used the same method, just with negative numbers. Figure 5 below shows my approximate defined change-no change threshold on the histogram.

(Figure 5)

After defining the change-no change threshold, I needed to apply that information in a visual manner. To do this, I used spatial modeler to create a model with the two input rasters from 1991 and 2011 and a function object to create an output raster. I viewed the output image's histogram and used the rule of thumb mean + 3 x standard deviation to find the change-no change threshold. After defining the threshold, I had to create another model (Consisting of one iput raster object, one function object, and one output raster object) that will show pixels that changed in the 1991-2011 study area. Finally, I draped the image showing change over an NIR band in order to show the change from 1991 to 2011 (Figure 6). Blue represents change. For the changes, It seems that there has been an increase in cropland. This could be due to forests being cut down or land being utilized as cropland that once wasn’t. There also seems to be lots of new developments near urban areas, which is not that surprising seeing as the city has definitely expanded in the span of that many years. The changes near urban areas seem to be due to urbanization while the areas for increased cropland could have to do with new farms or more forests being cut down.