Starting to put things together

I jumped back into work today after the memorable day off yesterday, and was ready to get some real work done on my project. I played around with classifications even more this morning trying to improve their accuracy, as always. I did some tests of my quantification process by using some trial imagery from Greg and Scott’s UAV common product. I did this by bringing everything into ArcMap, digitizing kuku’i stands, extracting kuku’i polygons from the eCognition file (which had over 11,000 unique polygons), georectifying the sample imagery, and more to get everything set up. Once features were digitized they were added as layers and shapefiles, and some geoprocessing calculations were completed. The first was the intersect of the digitized UAV visible regions of kuku’i, the second was the union of the eCog classification zones and the digitized UAV stands. Field geometry was utilized to calculate area for all polygons in all regions, and this was essential to measuring accuracy and quantity of overlap. The sum of the area of intersect was then divided by the sum of the area of union, and this calculates the coefficient of areal correlation for the classification of WV2 imagery. My initial assessment was low @ 31% however I believe this was attributed to improper/quick georeferencing, and an extremely small sample site (from initial X-8 captures). The second test I performed later in the day was of different X-8 imagery that covered a much larger region. Paul shared his already processed mosaic of this longer section, which offered a more stable reference base. This resulted in a quick, rough estimate of 61% overlap which was much better to see.

Tomorrow I will reorganize all my data to prevent future problems, and re-digitize and calculate the coefficient of areal correlation for the longer region. After this I will work with the new mosaic-ed Gatewing NIR imagery and hopefully I will be able to differentiate kuku’i on the landscape. I am also planning to utilize eCognition to perform feature extraction on my UAV imagery test sites, compare this to my digitized regions, and quantity the areal correlation for this object based classification as well. It will be interesting to compare how the software does between the two, however it must be noted that UAV imagery is 3-4 band whereas WV2 is 8 band. There will be much higher spatial resolution, hopefully not too high for the computer to process. I am excited to see the result of this and am really enjoying starting to put different elements together and quantify the product of the object based method. It can be frustrating at times working with software when you are trying to find the correct next step, but it is very satisfying to see things start to line up and make sense. Today was a great day and I am probably going to go to be early so I can go all-out tomorrow fixing problems and finding answers!