Remote Sensing Lab 4

Goals and Background

 In remote sensing there are multiple image functions that can be applied to improve the processing of aerial and satellite images. The goal of this lab was to learn these basic image functions. Some of the functions included the creation of a subset, improvement of image spatial resolution using pan sharpen, radiometric enhancement techniques, the use of a selective key to interpret images, resampling, image mosaicking, and binary change detection.

Methods

 Part 1: Creation of an Area of Interest (AOI)

There are two methods for creating an AOI: the inquire box and delineation with the use of a shapefile. The first method, the inquire box, is simply right clicking on the image and opening the inquire box. The box can be moved and resized as necessary. After the inquire box is correctly placed, the create subset image from the subset and chip tool will cut out the portion of the image in the inquire box and save it as a new file (figure 1).

Figure 1. Image showing the inquire box (white and black) around the Eau Claire and Chippewa Falls, WI region.

The second method, using a shapefile, is also relatively straight forward. A shapefile of the Eau Claire and Chippewa counties were added to the same image shown in figure 1. Both counties in the shapefile were selected and the 'paste from selected object' function was selected to create an AOI from the shapefile. This was then saved as an .aoi file.

Part 2: Image Fusion

In some instances, an image with a higher spatial resolution is needed for visual interpretation purposes. In this section the pan sharpen tool was used to accomplish this task. In a new viewer a reflective band image and a panchromatic band image were added. The reflective image had a spatial resolution of 30 meters and the panchromatic image had a spatial resolution of 15 meters. The resolution merge option within the pan sharpen tool was utilized. When merging the images, the multiplicative method and nearest neighbor resampling techniques were selected. The multiplicative method uses a multiplication algorithm to integrate the two raster images. The nearest neighbor sampling method determines the value of the new cell based on the the old cell value closest to its center. The final raster was the reflective image with the spatial resolution (15 meters) of the panchromatic image.

Part 3: Simple Radiometric Enhancement Techniques

There are several simple techniques that can improve the radiometric, or brightness value quality, of images. One such technique is haze reduction. Haze is a result of water in the atmosphere interfering with the reflected energy recorded by the sensor. The final image has clearer, more vibrant colors than the original without the haze.

Part 4: Linking Image Viewer to Google Earth

Within Erdas a user can sync an image in the viewer to Google Earth satellite imagery. Google Earth can then act as a selective key when interpreting images. Before interpretation, Google Earth is matched to the view shown in the image and the two views are synchronized (figure 2).

Figure 2. Google Earth satellite imagery synchronized to the spatial extent of the satellite image in Erdas.

Part 5: Resampling

Resampling is the process of making pixels larger or smaller than their original size. In this lab the pixel size was first resampled up from 30x30 to 15x15 meters using the nearest neighbor method. The process was then repeated using the bilinear interpolation method. The resample tool is located in the spatial category and is called 'resample pixel size.'

Part 6: Image Mosaicking 

In this portion of the lab two images captured in May 1995 by the Landsat TM satellite were mosaicked using both Mosaic Express and MosaicPro. The images cover path 25 row 29 and path 26 row 29. In both cases care was taken to make sure the higher quality image was placed on top. In MosaicPro, more properties must be specified before running the tool. Histogram matching was used to perform color corrections which ensures a smooth color transition from one image to the next. This was done only for the overlap areas.

Part 7: Binary Change Detection (Image Differencing)

 Binary change detection is a function that estimates and maps the change in pixel brightness values of a satellite image. Binary change detection for Eau Claire County and four other neighboring counties between August 1991 and August 2011 was completed using the 'two input operators' tool within 'two image functions.' The estimated cutoff for significantly different pixels was determined by adding and subtracting 1.5 times the standard deviation to the mean. The change in pixel brightness values was also mapped using spatial modeler. Instead of just subtracting the 1991 image from the 2011 image, a constant of 127 was added to eliminate all negative values. The final raster now has a change threshold solely in the upper tail of the histogram which is determined by adding 3 times the standard deviation to the mean. A new raster was then created using an 'either or' function to highlight pixels that changed using the change threshold calculated above.

Results

Figure 3. Subset of Eau Claire and Chippewa counties from a satellite image using a shapefile of Eau Claire and Chippewa counties.

The first task was fairly basic and straightforward. As mentioned in the methods, a shapefile of Eau Claire and Chippewa counties was added to a satellite image and used to create the subset in figure 3.

Figure 4. Pan sharpened image of Eau Claire and Chippewa Counties in Wisconsin.

After the completion of the pan sharpen tool the spatial resolution of the pan sharpened image shown in figure 4 was higher than the input reflective image. This is observed when the pan sharpened and input image are synced and one zooms in on both images. The input image will become pixelated sooner than the pan sharpened image.

Figure 5.Comparison between the input image (left) and the same image after haze reduction was applied (right).

In figure 5 the haze reduction image on the right has brighter pixel values compared to the input image on the left. The clearer pixel values creates higher contrast which makes demarcating objects in the image easier.

In part four of the lab an image was synced to Google Earth to aid in the interpretation of features in the satellite image. No outputs were created but the synced screens can be seen in figure 2.

Figure 6. Original input image (left) with 30 by 30 meter dimensions and the output image (right) resampled using the bilinear method to 15 by 15 meters.

In figure 6 one can see the smaller pixel size of the bilinearly resampled image compared to the original input image. In the resampled image the boundaries between objects appears more defined and distinguishable than the boundaries between objects in the original image. When the same image was resampled using the nearest neighbor method, the two images did not appear different. The pixel size and level of detail remained the same (figure 7). Although my images do not appear different, nearest neighbor should increase the stair step effect seen in linear features.

Figure 7. Original input image (left) with 30 by 30 meter dimensions and the output image (right) resampled using the nearest neighbor method to 15 by 15 meters.

Figure 8. Two sets of images mosaicked together. The two images on the left were mosaicked using Mosaic Express and the two images on the right were mosaicked using MosaicPro.

 Figure 8 shows the final output images after completing image mosaicking with two different functions. The image on the left was completed using Mosaic Express. This method is quick and requires no additional input but the final image does not appear as one image. The right part of the image is more red and brighter in color while the left side is more green and blue and duller in color. The boundary between the two original images is clear. The image on the left was created using MosaicPro. This method requires more computation time and inputs but makes a much better output image. MosaicPro synchronized the radiometric properties of each image along the boundary of the two original images using histogram matching. This resulted in a smoother transition between the two original images. Due to the poor quality of the output image created with Mosaic Express, Mosaic Pro is the recommended function to use.

Figure 9. Histogram of the data values for the binary change detection image created in part 7 of the lab. The dotted lines on both sides of the histogram mark the boundary between which pixels changed during 1991-2011 and which did not. The pixels that changed during the dates are located beyond the dotted lines towards the tails of the distribution.

The histogram shows the range of brightness values for the binary change image and the pixel values that changed from 1991-2011. The method of determining the boundary for change and no change was determined by the general rule of thumb that states the threshold is the mean plus 1.5 times the standard deviation. The positioning of these thresholds in figure 9 are an approximation. Binary change detection was also completed using spatial modeler. In this case a constant was added to eliminate all negative values and move all changed pixels to the upper tail of the histogram. A conditional function was then applied to show all pixels that changed and mask those that did not. The final result was mapped in ArcMap for easier viewing (figure 10).

Figure 10. Map of the pixels determined to have changed between 1991 and 2011 using binary change detection.

Using Google Earth as a reference, it appears that most of the changes over the 20 year period were somewhat near urban centers but were not due to urban sprawl. For example, there is a cluster of areas that changed near Lake Wissota and Chippewa Falls in the upper right of the image but the changed pixels are all located on farm fields. This same trend is true for the other clusters of changed pixels on the map.  Most of the changes depicted in the map were therefore due to a change in agriculture practices. Possible explanations could be a change in the type of crop grown which would change the reflected energy collected by the sensor or a change in soil quality.

Sources

 United States Geological Survey. (2018). Earth Resources Observation and Science Center [satellite images]. Retrieved from https://www.usgs.gov/science/mission-areas/climate-and-land-use-change/earth-resources-observation-and-science-center?qt-programs_l2_landing_page=0#qt-programs_l2_landing_page

Price, M. (2014). Mastering ArcGIS (6th ed.). Dubuque, Iowa: McGraw-Hill.