Module 5 - Supervised and Unsupervised Classification

    This week, in the fifth and final Photo Interpretation and Remote Sensing lab, it was about supervised and unsupervised classification. This lab is the last one for this course, as the next two weeks will be used for the final paper, however, the skills in this lab will be used extensively to finish the final paper. 

    The first part of the lab had us recode an image with unsupervised classification. This had us change the image first, so we could then manually select pixels and change them to one of five classifications. That being trees, grass, shadows, buildings or mixed. This took some time, and the lab showed us different ways to compare the image in ERDAS, as some pixels were mixed. An example of this was that some pixels on a roof were the same color as grass, these were made into mixed classification. Also, with each pixel changed, they were given an indicator for what classification they were. This was useful, as in the next step, we had to merge each pixel value with each other. Once done, we had a new image with only five layers for each classification. We did, however, have to manually recolor it, as the thematic change turned all the colors black. Once that was done, we used then calculated how much of the surface was permeable and impermeable, which rounded out the exercise.

    The second part of the lab was all about supervised classification. In it, we learned how to make our own signature figures in two different ways. One had us manually make the polygons, while the other had us use the inquire tool and grow tool in ERDAS to have the program make it for us. Once we had finished collecting all our signature figures, we then compared their histograms to check if there were any errors. In doing so we saw there were some with lower bands (1-3) but not the higher bands (4-6), this resulted in us changing the signature colors for the next step. In the last part of this section, we transformed the image, as well as attained a distance file. With the newly transformed image, we had reached where we were with recoding the image manually, but with the signature figures. The next step also had us combining the classifications like previously. Meanwhile the distance image showed where there were possible errors but was sufficient for the lab.

    The last part of the lab had us apply all we had learned to make our own supervised classification of Germantown, Maryland. As you can see below, I did my best on the classifications but may have had a small mixing of my road and building colors. Otherwise, I think I did well on the map. 


Comments

Post a Comment

Popular posts from this blog

About me

Hi, my name is Reese LeForce, and I wanted to share a bit about myself. I currently work for the Florida Department of Environmental Protection (FDEP) with concerns to Potable Water. I have also worked previously in the Stormwater department for a few years. I got into GIS to hopefully progress my career with the FDEP, and with the completion of the certificate, utilize GIS more within my work. Also, I may continue my education and get an environmental masters. Other than work, if I ever have the free time, I usually try and complete many of my ongoing projects of building model kits or finishing any games I have. I also try to make time to spend with friends and family whenever I can. I also have made a story map that goes a bit further in my past that I have linked  here . I hope you will engage with my blog and enjoy my GIS journey. 
Read more

Module 4 - Data Classification

Image
     This week, in the forth Computer Cartography lab, was about Data Classification.       In this module we used data classifications to differentiate our data. This involved using Equal Interval, Natural breaks, Quantile, and Standard Deviation classifications. With these classifications we are able to see the class distribution change with each map as well as learn what each one does. Equal Interval: groups data into equal grouped ranges that are uniform, however this provides errors due to masking clusters.  Natural breaks: groups data into natural groups where the data looks to have similarities in their clusters. Quantile: groups data similar to equal interval but equally distributes data into their category tracts, however this also conceals the data's true nature.  Standard Deviation: groups data on the deviation from the mean point of all of the data, this provides more errors if not normalized and also reads a bit harder.  ...
Read more

Module 1 - Fundamentals

Image
    This week, in the first Special Topics in Geographic Science lab, was about Fundamentals of data accuracy and precision.      In this classes lab we preformed analysis on a GPS data in order to determine how precise and accurate our data was to the reference points. As you can see in the map below we used the points given to make an average point which we used to make a precision buffer of 50, 68 and 90 percentiles. The 68 percentile is what we use for the majority of the lab but it does later lie close horizontally to our reference point.       For my horizontal accuracy, the distance was 3.25 m. For the horizontal precision, we got 4.4 m. Between these two sets of data we know that out data is accurate with as our reference points lies within out horizontal precision, but due to the rest of our waypoints being farther away from each other does mean our precision is lacking.      Why this is is because with accuracy, ...
Read more