Camera-based Water Stage and Discharge Prediction with Machine Learning

Objective

Find a model that predicts the stage and discharge of water in a body of water through the use of a numeric database and images of the site from the same angle over the years.

Database Content

The numeric database was divided into two types of data, generic and hand-crafted (57 independent variables). The second database contains all the images that were used to get the information for the former dataset.

Models

For the numeric database the implemented models were variations of:

• MLP Regressor
• KNN Regressor
• Random Forest Regressor

And for the images the implemented models were variations of:

• CNN
• MLP Regressor with Segmentation

Conclusion and Future Work

The best models consistenly were the CNN and versions of the MLP Reggresor, MLP Regressor with data from Segmentation having a slight edge over the others. To prove the complete effectiveness of the selected model it is theorized that it will require around 100 segmentation mask.

Development

Detailed information of the project can be found in the Research_paper.pdf file and implementation of the algorithms in the models directory.

Requirements

• A recent version of Python3 with pip to install libraries like TensorFlow and Keras
• Jupyter server environment