Remote Sensing - Module 5 - Unsupervised & Supervised Classification

This week's lab module for GIS 5027L Remote Sensing and Photo Interpretation focused on supervised vs unsupervised classification of images in ERDAS Imagine. The module introduced us to several tools such as swipe, flicker, and blend and concepts such as Spectral Euclidean Distance, Neighborhood, and Spectral Confusion. Some of the outcomes from completing this module included manually reclassifying and recoding images to simplify data, creating spectral signatures and AOI features, and recognizing and eliminating spectral confusion between spectral signatures.

The module began with a unsupervised classification exercise of an aerial photograph of the UWF campus. We learned various tools that help complete the process and learned about reclassification and how to merge classes by recoding. This part of the exercise ended with calculating the difference in surface types that occur in the image.

The second part of the module was an exercise in supervised classification which was much more in depth than the unsupervised classification exercise. Supervised classification involves Signature Collection where the user creates class type inputs, which are used to "train" the classifier to recognize features with different spectral characteristics. The user then evaluates the signatures to ensure that they accurately represent unique land covers leading to the most accurate classification. This is done by examining things such as histogram plots and mean plots looking for evidence of spectral confusion and spectral separation. After creating the supervised and distance images it is necessary to compare them to see if you notice any errors. To finish we merged all like classes and added an area column to calculate the area of the final classes.

In the final exercise in the module, we were tasked with creating a supervised classification of Germantown, Maryland.  In this exercise we employed all of the concepts that we learned throughout the lab module to create the final deliverable. According to the governor's office, over the past 30 years, Maryland's population has increased by 30 percent while land consumption has increased by 100 percent. The map below is a current land use map for the area in response to the Maryland governor's desire to work toward Maryland's "Smart, Green, and growing Initiative".

Remote Sensing - Module 4 - Spatial Enhancement & Multispectral Data Analysis

This week's lab module for Remote Sensing and Photo Interpretation introduced us to a wealth of information related to image enhancements and multispectral data analysis. This involved topics ranging from image enhancements in ERDAS and ArcGIS Pro to interpreting histograms and additional forms of digital data used to identify features. In the final exercise, we were provided criteria that were used to identify three different features. We employed the four main methods we had learned to identify these three features. The four main methods were as follows: 

1. Examine the histogram data for shapes and patterns in the data. 

2. Visually examine the image in grayscale for light and dark shapes and patterns. 

3. Visually examine the image with multispectral band combinations to isolate features of interest.

4. Use the Inquire Cursor to validate the exact brightness value of a feature.

Once we had identified our features according to the specified criteria we used the Create Subset Tool in ERDAS to extract an area around the feature allowing us to then export this subset into ArcGIS Pro to create a map layout. 

The first map below displays water features that were identified by a spike in band/layer 4 pixel values of 12-18. I decided to use the False Color IR as my band combination as it creates a sharp contrast between the water and vegetation.

The criteria for the second feature included a small spike in layers 1-4 around pixel value 200 and a large spike between pixel values 9 and 11 in layers 5 and 6. I looked at layers 1-4 using the panchromatic image to see the brightest areas. I proceeded to do the same for layers 5 and 6 to see what areas appeared darkest since pixel values between 9 and 11 would be very dark. After consulting the histograms and using the Inquire Cursor, I confirmed that the snow and ice features fit the criteria. I chose the False Natural Color as my band combination which displays the snow and ice in a light blue teal color.

The third and final criterion was to locate an example area that shows variations in water using a band combination that makes them stand out. Evidenced by layers 1-3 becoming much brighter than normal, layer 4 becoming somewhat brighter, and layers 5 and 6 remaining the same I identified the areas in the map below using a custom band combination displaying shallow and deep water.

Remote Sensing - Module 3 - Intro to ERDAS Imagine and Digital Data

The third lab for Remote Sensing and Photo Interpretation introduced us to ERDAS Imagine. This included familiarization with the layout of the workspace and the use of some of its basic tools and functions.  We began by learning how to properly add raster layers to the Viewer. We continued with an exercise adding both AVHRR (Advanced Very High Resolution Radiometer) imagery as well as Landsat Thematic Mapper (TM) imagery. This involved comparing the difference in spatial resolution between the two types. Using the Landsat image we changed the band combinations observing the differences between the different combinations such as False Natural Color vs False Color IR. It was very interesting to observe the visibility differences between the two. We concluded part A of the lab by creating a subset image of our original and added an area column in order to calculate the area of thematic classifications represented in our new subset image. We then opened our new subset image in ArcGIS Pro and created a map as seen below.

In part B of the lab we began by learning how to examine layer metadata through the Layer Info feature which contains important information about the image, including number of layers, file, image and pixel size, projection information and statistical information. Next was an exercise focusing on spatial resolution where we compared some images that depicted the same area on the ground but their level of detail varied. Next was a similar exercise about Radiometric Resolution and then we finished with an exercise with Thematic Rasters and Attributes. This involved calculating area and generating a formula statement in order to meet certain criteria.