RMBL_SnowView

Description

This is our project for the summer 2024 intership at RMBL. We are working with drone images of snow taken from 12 sites near Gothic, Colorado. The goal of the project is to develop a model that can perform image semantic segmentation on the images to mask where the snow is in the image. The goal of the project is to achieve a higher than 95% accuracy on the test set. We will also be making the model be abe to port to other datasets from other sites. It will also need to be used on future data from the same sites.

Iterations 1 through 6

Iteration 1 Jun 10

Aim to gain a better understanding of snow and images of snow

• Snow cycle near Gothic, CO and different areas
• Info of the location and surrounding area
• Other sites info

Iteration 2 Dataset

Aim to learn more about the dataset

• Location of the dataset
• Pre-processing steps
• Data cleaning steps
• How balanced is the dataset
• How to split the dataset
• are there any bad images/masks
• how to quantify the results

Iteration 3 Exploration of Models

Aim to learn more about possible models that we will use

• Models:

  1. CNN
  2. RF
  3. Auto Encoder
  4. SVM

• Results of the models

• Are there any better pre-trained models we should use or should we train our own?

Iteration 4 Model Building

Planning out our model and beginning to build it

• Analyze Iteration 2 results
• Create Plan for model
• Create resources requirments
• Create responsibility plan
• Create protocols
• Diagram the model
• Code and document

Iteration 5 Model Tuning and Testing

Tune the model understand the results

• Tune:
  • Hyperparameters and other parameters
  • Structure
  • Pre-processing
• Visualize and store results

Iteration 6 Culmination

Finalize the model, prepare for submission and presentation

• Overview final presentation
• Write documentation draft/submission
• Finalize code
• Submit

Contributors

• Andrew Fox
• Joel Carlson