DeepRivWidth

Deep Learning based Semantic Segmentation approach for River Identification and Width Measurement in SAR Images of Coastal Karnataka

Abstract

The manuscipt is currently under peer-review. The methodology and the results will be updated soon!

Download

This dataset is made available for academic research purpose only, and should not be used for any commercial purpose. To download the dataset, please fill the Google Form. The link to download the dataset shall be shared to your email address.

A sample SAR image along with the annotated ground truth is shown below:

DeepWidth Code Base

The Directory of Deepwidth is arranged as

This setup currently works on single images only. This project has 2 functionalities : Segmentation and width measurement.

• The graphs of training are in the plots folder.
• The saved models/weights are stored in the weights folder.

Usage Instructions

Segmentation

• Place SAR image in the input_imgs folder.
• Run main.py in the src folder.
• Choose model you want to test by giving using 1 for DeepLabv3+ and 2 for UNet.
• Final image will be stored in the output_imgs folder. Intermediate images are stored in the sub_input_imgs and sub_output_imgs folders for inspection of each sub-image (256x256 crop)

Width Measurement

• [Optional] Run the set_scale.py script to set the scale. scale.png has the scale information. Press s to enter set scale mode. Click and hold and drag mouse till wherever you want to measure and let go of the click. Scale information would be presented at the console.
• Edit the scale variable in the distance_measurement.py if scale has changed and run.
• Press c to select region of which you want to measure width of. The distance (in Km) will be returned on the console.