Skip to main content

Water Level Estimation

Paper​

  • Paper by Haurum et al (2020) on Water level Estimation in Sewer Pipes Using Deep Convolutional Neural Networks

Sudhir's Comment​

  • Received on 10-03-2022. Subject: I read the "Water Level Classification" paper by Haurum that Vannary had sent
    • I think Vannary should be able to use the existing data and formulate a multi-label classification problem for water level estimation using the WL class information that is present in the CSV.So only the CSV will have to be reformatted (drop all defects information) and then our FastAI model can be used.

Notes:​

  • When the inspection occurs? if they do it after storm event. Subtract with the dry flow to get I&I. Delay
  • Water segmentation: use flood segmentation for water level. Sudhir is thinking of leverage the flood segment technique to do the water level
  • One hot encoding vs label for water level interval (labeling ordernal class)
  • Ask Joakim, how does CSV has 1%, 2%... class? Waiting for his reply
  • Write python script to divide water level into class (0-10 -> label as 10 class) Single label classification.
  • If they inspect after the storm, we can we use this to estimate the I&I
  • CSV file Pavan from json fname and two character labels. (distance, dates, clock angle)

Train water estimation model​

  • The water level is assigned based on how full the pipe is in percentage for both Fotomanual and NASSCO PACP. The image from SewerML can be used to help train DNV water level estimation model.

Proposed run for water level estimation model​

  1. Get the distribution of water level of SewerML dataset
  2. Get the distribution of water level of SewerML dataset only on images that have VA as label
  3. Compare point 1 and 2
  4. At this model, I only submit model
Last updated on by jacobgqc