Project 1: Labeling and Geological Interpretation of Martian Orbiter Photos
- Goal: Label Lineaments and Cracks in Mars Orbiter Pictures using Transfer Learning and the Sandstone U-Net. Read U-Net Crack Detection in Photos of Sandstone and Volcanic Rocks.
- Procedure: Download Shi's Keras lab with the U-Net sandstone architecture and trained U-Net weights. Download
Mars orbiter pictures from NASA or ESA pictures. Contact GTS if you are interested and plan to put serious time into project. I will then get you in contact
with renowned geologists Margie Chan (marjorie.chan@utah.edu) or Ron Bruhn (ronbruhn46@gmail.com), who have been field oriented geologists for more than 35 years. Professor Chan has been involved in using sedimentary analogues on earth to reveal the geology of mars, and was one of the first to show that the blueberry rocks in Southern Utah are the same as those discovered by a Martian lander on Mars. Ron is one of the best structural geologists
in the world. They will give you geology papers to read so that your project will try to solve a geology problem on Mars rather than just mapping a bunch of lineaments.
You can practice with transfer leraning and crack detection while you study the papers and define the geology problem you will solve.
- Step 1: Define geology problem and download appropriate images.
- Step 2: Transfer learning from sandstone U-Net and modest amount of extra labeling effort to get a hybrid U-Net.
- Step 3: Label data and perform interpretation with assistance of above geologists.
Project 2: Labeling and Geological Interpretation of Martian Rover Photos
- Goal: Label Lineaments and Cracks in Mars Rover Pictures using Transfer Learning and the Sandstone U-Net. Same as Project 1 except use the Martian Curioisty Rover pictures at NASA.
Project 3: Labeling and Geological Interpretation of Planetary Photos
- Goal: Same as Project 2, except label Lineaments and Cracks in Planetary Pictures
using Transfer Learning and the Sandstone U-Net. Same as Project 1 & 2 except use NASA's planetary pictures of
Pluto or Enceladus. About four years ago at WEP, I hosted the imaging team manager (Carol Porco) of the Cassini mission that imaged Enceladus, so if you have a strong interest in doing a project on Enceladus I can contact her and perhaps we can have a NASA expert guide you. This requires a huge time commitment and follow-through. No superficial commitments.
Project 4: Labeling Bird Pictures from GTS Backyard
- Goal: Label Bird Pictures from GTS Backyard and Around Campus.
- Step 1: Contact Keras manager of the bird project Yunsong Huang at yunsongh@gmail.com.
- Step 2: Download bird pictures here.
- Step 3: Label bird pictures and train Keras code for Alexnet. Unlike the Projects 1-4, this is not semantic segmentation CNN.
- Step 4: Apply trained CNN to test data.
- Step 5: Download CNN to iphone and have iphone camera identify birds. Will show you site for download app for Iphone. (ask Yunsong Huang and Dan Trentman for assistance).
Project 5: Labeling and Counting Fish Pictures from Ron Bruhn
- Goal: Same as Project 4 except Label Fish Pictures from Ron Bruhn. Many big dams on Columbia Rver need to count fish going up fish ladders. Develop a ML method to do this.
- Step 1: Download fish movie here.
- Step 2: Decompose movie into pictures (ask Yunsong Huang for assistance). Two types of fish: perch and salmon.
- Step 3: Follow steps in project 4.
Project 6: Labeling and Counting Bomas in Kenya
- Goal: Same as Project 4 except label and count bomas in GoogleEarth pictures from Kenya. Bomas are enclosures used by herdsmen to protect themselves from wild
animals at night. Archaelogists and socialogists are very interested in their distribution in Kenya. See me if interested. One of the world's most renowned geochemist-geologists might be willing to work with you.
Alternatively, one of the Leakey's in Kenya is working on an eb=nvironmental project that could probbaly use ML. here4 is an email one of them wrote in response to possible projects: "Thank you very much for considering me for this opportunity! I'm really interested in machine learning but will not be able to take this course at the moment as we are about to start a new project in Ileret looking at energy, water and agriculture." Read paper to get a background on some environmental issues in Kenya.
Project 7: Jurassic Navajo Sandstone, SE Utah (Eolian Wind blown sandstone)
- Normal cross bedded dune sandstone (angles up to 34 degrees angle of repose), usually shows a pretty strong anisotropy (often weathers as ribs of sandstone). Some major bedding surfaces and interdune deposits can be horizontal.
- Deformed sandstone; swirly patterns, typically show some angles of internal bedding exceeding 34 degrees. The most deformed is massive where nearly all original bedding is gone. This shows the smoothest weathering pattern. See PPT description. Margie Chan is in charge of this project.
Goals:
1). Develop ML to distinguish normal cross bedded sandstone from soft sediment deformation (SSD) where cross bedding was disrupted and disturbed from liquefaction and fluidization (original textures were altered probably due to triggers like equakes.Use the sandstone U-Net weights and Keras code. Read DCNN as a geological image classififcation tool. Download the zip file of all the Utah pictures.
2). Apply ML to map Utah examples as well as in remote imagery from Mars.