GIS4035 - Module 6 Lab - Spatial Enhancement

In this first lab for Spatial Enhancements we learned various filtering methods that can be used to enhance an image depending on the analysis and interpretation needs.   The different exercises and tasks demonstrated how to use a variety of methods both in ERDAS and in ArcMap.

Our First Exercise required us to learn how to download imagery data.  We were provided a USGS website and a specific sensor.   We changed the location to center the viewer on Pensacola Bay and selected a specified image date (Julian Date).   The name of the image is a description of the sensor type, location and date of the data.  Once downloaded the data needs to be imported into ERDAS and converted into an image file.

In the Second Exercise ERDAS Convolution tools under the Raster, Spatial menu, were used to create two basic enhanced images; a low pass filtered image and a high pass filtered image.  Low pass filters “smooth” out the data by eliminating erroneous pixels or unnecessary small features.  The high pass kernel type highlights small scale differences or high levels of changes in cell values from cell to cell, eliminating broad patterns but highlighting “edges” or feature types such as roads or water bodies.

We also used Focal Statistics in ArcMap to apply statistical filters to an image.  Focal Stats performs various statistical tests using an adjustable kernel of various shapes and sizes.  New values are calculated for each cell in the raster.   The lab had us create a Rectangular Mean 7x7 filter which is a low pass filter.  A larger kernel size results in each new cell value being the average of a larger area so the output image will be more generalized.  The second filter we applied to the original image was an Edge Detect filter using the Range statistic.  This filter type gives each new cell a value that reflects the difference in brightness between the neighboring pixels which can help reveal the borders between different types of features. 

In Exercise Three we used ERDAS to perform Fourier Analysis to diminish the effects of striping bands that can appear in an image due to sensor malfunction which causes frequent lines of no data.  This method is under the Raster tab, Scientific group, Fourier Analysis button.  You first create a Fourier Transform file and then use it as the input to create an image file.  The Fourier Transform Editor has various tool buttons that can be used to enhance the image.   Low and High Pass filters and various Masks.   We used the Wedge Mask and Low Pass filter to diminish the striping effect.

We further enhanced the image by using the Sharpen tool in the Convolution toolset.  The Fourier image file previously created was used as the input file.  The 3x3 Sharpen filter is a high pass filter that sharpens edges but does not reduce overall details in the image.  Different kernel methods can be used to filter images.  I found this filter selection guide that provides a description for various filters.

https://hexagongeospatial.fluidtopics.net/book#!book;uri=5d68e1db557af5bab494d96c7f8e1a9d;breadcrumb=98d11f6e256a86a22eaebb041a3f6075-e2158ff6602259c9651bd3ed7f08262e-5b95c34a774882c13680195f5cc2e0fb-0d93bff57e8bbede8d9c7bd229e8a569

My final deliverable, Task 3 for Exercise 3 started from scratch using the original striped image (l7_striping.img).  First, I performed a Fourier Transform on the image.   This was my third time using this analysis technique in ERDAS and I was able to create a better fft file than my first two attempts from the lab because I utilized the Wedge and Low Pass buttons more effectively when dragging the cursor.

My output Fourier3.fft was used to create my Fourier3.img file.  This Fourier image was used as the input file for Convolution Filtering.   I first applied an Edge Enhanced 3x3 Kernel and created edgen3x3_3.img.  I read using the ERDAS Help that Edge Enhance brings out the edges between homogeneous groups of pixels, unlike edge detectors (such as zero-sum kernels), they highlight edges and do not necessarily eliminate other features.  (ERDAS)  I didn’t want to lose any features in the image so I felt this was a good kernel to start with.  When I compared this Edge Enhanced image to the original, it appeared a little fuzzy, so I chose to apply another Convolution Filter, Sharpen 3x3 to slightly sharpen some of the edges (sharpedgeen3.img).  The northwest corner to the bottom of this sharpened image had a very dark appearance and I needed to “subtract” some darkness from those objects/features.   I read up on using the Breakpoints Editor in ERDAS but did not feel comfortable using it to adjust the image.   Instead, I chose to use the Brightness Contrast under the Panchromatic tab to brighten the final image.  I felt this enabled the various features in the image to be more visible.

Once I opened the final enhanced image (brisharpenedgeen3.img) in ArcMap I wanted to show the original image contrasted with the enhanced image.   I chose to display three data frames.   The first data frame displays the full extent enhanced image as required for the lab deliverable.   The other two frames are focused on a section of the image I referred to as “mama bear”.  This side by side comparison I felt showed the effects of the different enhancement techniques I used on the original image. I had difficulty displaying the original image in both Data and Layout views.   I could not display the full extent of the original image.   In retrospect I should have shown the “study area” on the full extent frame of the enhanced image but I did not go back and try to mess with it. 

This was another challenging lab. I vaguely remember using the Fourier Theory in Electrical Engineering courses but it was so long ago and that knowledge was never applied during my career experiences.  There clearly is so much material involved in remote sensing and we are just scanning the surface in these labs. 

Image Enhancement of Pensacola Bay Using Fourier Analysis and Convolution Tools in ERDAS

GIS4035 - Module5a - Intro to ERDAS Imagine and Digital Data 1

This week's lab appeared simple enough but truly understanding the differences in images and the devices used to capture them as well as continuing to learn about the properties of Electromagnetic Radiation and the Electromagnetic Spectrum proved to be challenging.  This class so far has covered more subject matter material than previous courses.   The focus points of this lab were to ensure we knew how to calculate the wavelength, frequency, and energy of Electromagnetic Radiation and to learn about and how to use ERDAS Imagine to view and create images.

The first exercise was to perform simple calculations using formulas such as Maxwell's Wave Theory: The speed of light is equal to the wavelength in meters times the frequency in Hertz.  The important takeaway was the inverse relationship between wavelength and frequency.  The longer the wavelength the lower the frequency and conversely, the shorter the wavelength the higher the frequency.  The wavelength can be used to determine the spectral region/location.  

Max Planck proposed the quantum theory of electromagnetic radiation where energy is transferred in discrete packets called photons (Jensen, p. 45).   The energy Q, of a quantum measured in joules, is equal to Planck's constant multiplied by the frequency of the radiation.  The relationship between energy and wavelength is also important.  Longer wavelengths result in lower energy content.

The second exercise introduced us to ERDAS Imagine.  We learned how to open provided raster images of different types and how to display the images using the Raster Options tab.   We used different tools and techniques to view the images such as panning, zooming, fitting the image to frame or changing its scale as well as how to display multiple images.  We also viewed an Arc Coverage file which is a digital vector format file that can be found in the ERDAS Imagine default installation filepath.   The images we viewed were obtained from different devices.   One image was from an Advanced Very High Resolution Radiometer (AVHRR) and the other images were obtained using a Landsat Thematic Mapper (TM).   These images had vastly different spatial resolutions.   This exercise also had us explore the different band combinations for multispectral data.  Changing the band combination can aid in identifying different elements in the image.

The final exercise had us open a thematic single layer in ERDAS Imagine and tasked us to focus on a smaller area of the image.   Using ERDAS we were able to create a subset image which was then used to create a map in ArcMap.   While in ERDAS we also viewed the image's attribute table and added a new "Area" column that calculated the area of each classification.

Classified Subset Image of Washington State Created in ERDAS Imagine

The subject material in this lab was the challenge for me this week; understanding the concepts and properties of EMR.  Learning how to change the view or display of images, in order to aid in the interpretation of feature elements, was the fun part.

GIS 4035 - Module 4 Lab - Ground Truthing and Accuracy

In this week's assignment we were tasked to "ground truth" our Lab 3 map.   In lieu of "in situ" data collection, we used Google Maps to verify the classification codes we assigned to different areas on the image file in the Lab 3 map.  

1.   First I opened ArcCatalog and created a Truthing point shapefile with the same projection as the TIFF image and LULC shapefile from Lab 3.

2.   Next I opened up a copy of the Lab 3 map and added the Truthing layer and saved the new Lab 4 map.

3.   Next I reviewed the map and used sampling rules of thumb to determine my sampling requirements.

a.     Lab 3 was classified to Level II so sampling the transportation class was probably not important.

b.     Bodies of water and Level II wetlands I assumed were homogenous and decided not to locate samples in these classes.

c.     The large coverage area of residential classification (11) led me to choose the stratified random sampling method since this method is proportional to LULC type.

d.     I placed at least 1 sample in each classification code (except Water, Wetlands, and Transportation, see above).

e.     I added samples proportionately to each classification based on the coverage area size.

4.   I began creating point features in the Truthing shapefile after adding the required fields.

5.   I then added a field for coordinates of each point feature.  Using the Identify tool I recorded the degrees, minutes, and seconds of each point.

6.   Next, I opened Google Maps and began the ground truthing process using the coordinates of each point location.   I took notes in a comment field I had added so I wouldn’t get confused between points.  

7.   As part of this lab I realized that I did not maintain scale or MMU properly in Lab 3.  I tried to keep my scale to 1:10,000 in Data View when randomly locating my sample points.   Likewise, in Google Maps I tried to be aware of the scale in the bottom right before going into Street View.   I think if I had consistently applied MMU and scale in Lab 3 I would have had fewer classifications.   Fewer classifications may have improved my accuracy in Lab 4.

8.     The biggest issue for me in classification was that there were several “mixed use” areas on the map.   For example, while some areas would appear predominately residential my sampling point was on a large commercial building that was “mixed in” within residential.  Seven of the eight misclassified sample points should have been classified as either Residential (11) or Commercial and Services (12). I think this might indicate lax zoning ordinances and poor urban planning.

Ground Truthing of Lab 3 Map Using 30 Stratified Random Sample Points and Google Maps

GIS4035 - Module 3 Lab - Land Use/Land Cover Classification

In this week's lab we continued learning how to interpret aerial photographs using various elements but took it another step further by then classifying the interpreted areas using the USGS Land Use/Land Cover Classification System.  This module again covered a large amount of information.   Wrapping my head around the text and supplemental readings was overwhelming.  Once I started the lab however I was able to let go of some of the details and depth of information and focus on a Level II classification of the provided image.   This one image assignment made it clear how enormous a task digitizing an image can be and the amount of detail that could be required.   As the lab assignment stated "we could spend all semester examining this one photograph."

Lab 3 - LULC Classification Of Features In An  Image From Mississippi

1.     First step was to open ArcCatalog and review the TIFF image that was provided for the lab assignment.  

2.     Second, was to create a new shapefile, also in ArcCatalog, and then to set its coordinate system to match that of the TIFF file.

3.     The previous week’s lab had us use the Drawing toolbar to create polygons and then convert the graphics to a feature shapefile.   This week we used the Editing toolbar to build the polygon features that would be contained in the shapefile. 

4.     I recalled only using the Editing toolbar one other time in a class last Spring so I had to refresh my snapping and vertices techniques as well as learn about the Trace option.

5.     Saving edits was the key not only after creating the feature but also after editing the attribute table for each feature.

6.     I used the clip option for several of the polygons where there was overlap or intersection.  I feel the clip option could have been better utilized when initially creating a plan on digitizing the various classes.   Since the residential and body of water sections were a majority of land use/land cover in this image, in retrospect I probably should have created those class polygons first.

7.     Because I did very little mapping in my last class (Python programming), I am having to get back into the mapping mode.   Working on this map legend helped refresh some of my skills.

8.     I figured out how to display the spatial resolution on the TOC after using ArcGIS Help for spatial resolution.   I did not figure out the best way to calculate or formulate a MMU (Minimum Mapping Unit) for this image.   I roughly chose an MMU of 1/8 of a mile or somewhere between 500-660 ft².