Spectral Classification

For this week's lab, we learned how to complete an unsupervised and supervised classification on an aerial image in ERDAS Imagine. First, we practiced classifying an unsupervised image of UWF campus by accurately classifying different spatial and spectral resolutions. To finalize this, we reclassified and recoded images from the spectral signatures created in an AOI layer. 

After learning the steps needed to create a supervised classification map, I used the inquire tool to enter in the coordinates for all the features to begin creating spectral signatures (except for the roads and water-I created them without the inquire tool). When I completed creating the spectral signatures, I evaluated them using the Histogram and mean plots. The bands with the greatest distance from each other determined the color band that was used. For this map, I used R4, 5G, 3B. The output distance file did show some bright pixels that let me know the classification was not correct. I had to go back to my spectral signatures and adjust the Fallow, Water, and Road area. I could never get the Road and Urban area perfect but I finally got an image that made sense.

Spatial Enhancements & Multispectral Bands

For this week's lab, we learned how to apply spatial enhancements, perform a multispectral analysis, and create band indices.

The spike on the histogram between pixels 12 and 18 is very tall so that means the feature is very large. I viewed in multiple color combinations and greyscale and noted features I was confident with identifying such as water, mountains, town. When I moved the breakpoint on the histogram between 12 and 18 pixels, the rivers and lakes remained dark so I was able to conclude that the feature was water. For the bands, I used the TM False Natural Color 5,4,3 Landstat 4 TM-6 Bands because this combination band makes water delineation stand out.

I viewed this image in greyscale and different multispectral bands and it was white in Layers 1-4 and dark in Layers 5-6. When I viewed the Histogram, there was a small spike on the right which means it is very bright.  I was able to identify the feature as snow. I used the True Color combination 3,2,1 with the Landstat 4 TM-6 bands because the white snow stood out in natural color. 

Since I knew this feature was going to be water as well, I just had to deduce what type of water. Layers 1-3 became brighter in certain areas and somewhat brighter in certain areas in Layer 4. I was able to quickly identify the feature as being shallow water because the shorelines were brighter and then what appeared to be a landlocked water basin was very bright as well.  I chose the TM False Natural Color IR 4,3,2 combination with Landstat 4 TM-6 bands.

ERDAS Imagine

 For Module 4, we were introduced to ERDAS Imagine and how to use basic tools and view data. This lab really jump started our work with remote sensing. We learned about EMR, electromagnetic radiation, and how to calculate its wavelength and frequency based on Maxwell's Wave Theory. The size of the wavelength is related to the amount of energy and the frequency of that energy. The amount of energy is inversely related to the wavelength of the light: the shorter the wavelength, the greater the energy of each photon of the light. Maxwell's Wave Theory, which is the speed of light = (wavelength in meters)*(frequency in Hz), and Planck's Relation, energy of photon/quantam in Joules = (Planck's constant- 6.626*10^-34 J*s)*(frequency in Hz), calculate and derive variables that explain EMR. This is important to understand when studying remote sensing because it helps you identify features in the photography. In ERDAS, we learned how to view raster layers and understand the data being displayed. Raster layers can be raw/multiple layer continuous data, single layer panchromatic continuous data, or categorical/single layer thematic data. Continuous data is quantitative data that has related, continuous data. Thematic data is qualitative data that is categorical. We compared AVHRR images to TM (thematic mapping) images to determine that the former has a coarser spatial resolution than the latter. On the TM image, many features can be identified, so we changed the band combinations for multispectral data to enhance vegetation differentiation, snow, water delineation, and bare ground/urban areas. We also learned that to switch between the TM False Natural Color and TM False Color IR, the former makes the bare ground more distinguishable. These are important tools to use when identifying features. For the map delivarable this week, we used ERDAS Imagine to create a subset image to use in ArcPro to make a map.

Land Use/Land Cover Classification and Ground Truthing

For Module 2, we utilized the identifying features we learned last week, such as shape/size, color, shadow, pattern, and association, to help locate features in a natural color aerial image and create a land cover/land use map in Pascagoula, MS. Land cover is the biophysical description of the Earth's surface. Land use is how humans interact with the landscape and therefore change it. I used Level I and Level II classification scheme to create codes for different features I found. There are 9 Level I classes and I used 6 of them: Water, Urban, Agricultural, Wetland, Barren Land, and Forest Land. The trickiest part of this lab was getting comfortable creating polygons and editing the vertices in a way to keep overlap at a minimal.For the ground truthing calculation, I created 30 random sample points by distributing them as evenly as possible across the map. Then, I utilized Google Maps street view to check the accuracy of each point to polygon. In the shapefile attribute table, I used Yes or No to confirm the accuracy. 

Visual Interpretation

For exercise 1, I had to learn how to identify features using tone and texture. I began with “tone” which is how light or dark something is, and created a feature class for each classification. This is done by highlighting the .gdb file in the ArcCatalog pane, right-clicking, and selecting new feature class with polygon geometry. I added the Tone feature class to the map by dragging it over. Then, I added each feature name opening the attribute table for Tone, and adding a field for each feature. Lastly, I would create the polygon for each feature by highlighting the feature in the attribute table, selecting edit>create feature>polygon. I repeated this process for Texture For labeling purposes, I chose the polygon lines to be a bright blue color (Tone) and a bright red color (Texture) with a 2pt width because I needed it to stand out on an otherwise colorless background. The 5 classifications I used for Tone were Very Light, Light, Medium, Dark, and Very Dark. The 5 classifications I used for Texture were Very Fine, Fine, Mottled, Coarse, and Very Coarse. Some examples of features I chose for tone and texture were an airport for very light, calm water for very fine, and a neighborhood for coarse. Some features had varying tones/textures such as vegetation.

 For exercise 2, I had to visually identify features using the criteria of shape/size, shadow, pattern, and association. I used the same steps as in exercise 1 to create feature classes for each criteria except this time I used point geometry. Shape/size is the most intuitive and easiest identifying feature to begin with. Without having to zoom in or study the aerial photo long, I was able to identify a pier. I was pretty sure I located a two swimming pools but after zooming in slightly, I was 100% sure. I did have to zoom in very close and pan around the photo to see other features more clearly. This allowed me to easily identify a vehicle. I remained zoomed in when I chose the three features for shadow. I chose trees, a motel sign, and a water tower because they had clear shadows being cast. I almost chose buildings because so many had shadows but decided to use the examples that would be very difficult to identify without that one criteria (buildings have multiple criteria). Pattern is almost as intuitive as shape/size for identifying features, in my opinion. It was easy to locate the waves in the body of water, the multiple transmission lines, and the lined spaces for the parking lot. For association, I chose the motel and the neighborhood clubhouse. Each building contained a swimming pool which helped signal the association. For labeling purposes, I employed a few techniques for the map to stand out, as I was dealing with another colorless background. I used 5pt circles with different colors for each criteria. For the labels, I used the Point of Interest text symbol with “white halo”. However, I changed the color of the halo to match the symbol color, used with the underlying black text. I felt this aesthetic was pleasing to the eye. 

For Exercise 3, I explored color interpretation by comparing a true color image and a color infrared image. The mixed pine forest area southwest of the area initially appeared green to me but it was entirely red in the false color image thus its true color was near infrared. The marsh was easy to discern as its true color was green and false color was blue. There were some areas that I interpreted as rivers/water retention areas that were white in true color and blue in false color. I concluded that these could be possible construction or industrial areas that do not contain runoff all of the time. The buildings appeared to be white in both true color and false color because they reflect light. We did not have to make a map of this exercise but it was good practice to see how having a thorough understanding of your area and how true/false color can help you understand features in an aerial photograph.