User Tools

Site Tools


atmos:citation:research:test_classification_tar_file

Test PHIPS Classification Dataset for the 20190803_1424 flight during CapeEx19

2021/10/4

Christian Nairy christian.nairy@und.edu

############# SUMMARY ###############

Due to the non-homogeneity of observed chain aggregates, a new definition was developed for my thesis. The new definition is not Boolean in nature but can range in confidence. In this work a chain aggregate is defined by having at least one of the following characteristics:

(1) - Three or more discernible particles oriented in a quasi-linear fashion. (2) - Particles joined together by small joints. (3) - Links of particles that are unusually elongated.

If one, two, or all three definitions are observed, a confidence value of one, two, or three, is applied respectively; where confidence-one is low confidence, confidence-two is moderate confidence, and confidence-three is high confidence.

############# OBJECTIVE ###############

The *objective* is for you go through the PHIPS images provided and classify the images as a chain aggregate or not a chain aggregate and give it a confidence value based on a reference key that is given to you.

I will then use your classification dataset as well as other datasets performed by others and provide statistics on the agreements/disagreements.

Thank you for going through these PHIPS images. This will really help me with my thesis project.

############# INSTRUCTIONS: ###########

(1) Download this file: test_classification.tar.gz

(2) You will need to open the contents of this folder. If you are using linux, use the command line where the file is located and type:

      tar -zxvf test_classification.tar.gz

In this folder, a csv file (Test_classification_wksht.csv), a reference key (chain_agg_reference_key.pdf), a folder with 100 PHIPS images (20190803-1424_C1-C2) should appear.

The reference key PDF contains examples of the chain aggregates with varying confidence and their respected definitions. You will use this when classifying the PHIPS images.

In the PHIPS images folder, you'll see '.png' files (e.g., PhipsData_20190803-1424_004464_15:51:17.273.C1-C2.png). The naming convection is not important, though, the PHIPS image number in this example is 4464.

In each .png image, there are two images (C1-C2) of the same particle. The C1 and C2 cameras are offset by 120 degrees. You can use the two camera views to your advantage when classifying the images.

(3) Now, open up the CSV file. This will be your worksheet while classifying the PHIPS images.

The left column is the PHIPS image number column. You won't need to do any manipulation in this column.

The middle column is the 'Chain Aggregate?' column. While going through the images, if it's a chain aggregate insert a '1', if it's not a chain aggregate insert a '0'. See definitions above and/or the chain aggregate reference key PDF provided.

The right column is the 'Confidence' column. This is where you will use the chain aggregate reference key for your decision. If you classify the chain aggregate as confidence-one insert '1', confidence-two insert '2', and confidence-three insert '3'. If the image that you classified is NOT a chain aggregate, insert a '0'.

(4) Next, open the PHIPS images and begin your classification. Make sure you reference back to the chain aggregate reference key! Also, make sure the image you are classifying matches to the image number in the .CSV file!

(5) When finished, please save the file as a .csv file then send the saved file to my email christian.nairy@und.edu

atmos/citation/research/test_classification_tar_file.txt · Last modified: 2021/10/04 18:17 by 127.0.0.1