Accurate Water Feature Mapping Using Deep Learning

Accurate Water Feature Mapping Using Deep Learning

Table of Contents:

1. Introduction
2. Purpose of the Project
3. Water Feature Mapping
4. Models for Water Feature Extraction -Unit Model -Encoder/Decoder Model
5. Data Set and Model Evaluation
6. Training and Testing the Models
7. Results
8. Performance Evaluation Metrics
9. Advantages of Water Feature Mapping
10. Future Development

Water Feature Mapping: A Deep Learning Approach for Accurate Extraction of Water Features

Water feature mapping is a crucial process to ensure access to safe and reliable water resources for various purposes such as irrigation, flood forecasting, and drought management. With the use of satellite imagery, it has become possible to map and extract water features accurately. In this article, we will discuss the process of water feature mapping using deep learning techniques and how it can revolutionize the way we manage our water resources.

Purpose of the Project

The main purpose of our project is to extract linear and nonlinear water features from the Sentinel-2 satellite imagery using a unique deep learning architecture. We aim to extract various water objects from medium-resolution imagery by utilizing deep learning approaches. Our team, including our mentor, Dr. Philip Terraces, and teammates at the Pennsylvania University, are working on developing and training two models: the Unit Model and the Encoder/Decoder Model.

Water Feature Mapping

Water feature mapping involves the process of extraction and mapping of surface water features such as rivers, streams, lakes, and others. Mapping water features play a vital role in providing water services to urban consumers, agricultural areas, and forecasting floods and droughts. With the adoption of deep learning techniques, it has become possible to map the water features with higher accuracy.

Models for Water Feature Extraction

Our project involves the development of two models: the Unit Model and Encoder/Decoder Model. The Unit Model consists of convolutional and max-pooling layers in encoding and decoding layers. We have made some modifications to make this model suitable for our application. The Encoder/Decoder Model, on the other HAND, consists of convolutional, max-pooling, and batch normalization layers in encoding layers and upsampling and conv2dtranspose in decoding layers. We have also made some modifications to this model to make it more suitable for our application.

Data Set and Model Evaluation

For our project, we have used a data set consisting of 2160 images with a pixel Height and width of 4505x5062 and a resolution of 10 meters. We have divided the data set into training, testing, and validation sets. We trained both the Unit Model and the Encoder/Decoder Model for 200 epochs using the training and validation set. We evaluated the models using the testing set.

Training and Testing the Models

We extracted the ground truth images and the images from the data set into arrays and augmented the data. We then prepared the models and trained them for 200 epochs. We tested the models using the testing set, and the accuracy, loss, and prediction outputs of the models are shown in the result section.

Results

We were not able to get the desired result using the Unit Model, so we implemented the Encoder/Decoder Model. We were able to map the water more accurately using this model. We achieved an accuracy score of 99.96%, compared to the accuracy score of 99.93% in the Unit Model. The precision score was 88.11, recall was 90.75, and F1 score was 89.41.

Performance Evaluation Metrics

The performance of our model was evaluated using various metrics such as accuracy, precision, recall, and F1 score. We achieved an accuracy score of 99.96%, which reflects the percentage of correctly classified samples. The precision score was 88.11%, reflecting the number of true positives among the total predicted positives. The recall score was 90.75%, which is the ratio of true positive to all actual positive samples. And finally, the F1 score was 89.41%, which is a combination of precision and recall.

Advantages of Water Feature Mapping

Water feature mapping is essential for smart GRID and other organization developments. It can help with the management of water resources, disaster management, agricultural monitoring, and much more. With the use of deep learning techniques, it has become possible to map the water features accurately, enabling us to make informed decisions about water management.

Future Development

In the future, this model can be used to extract other features such as agricultural areas, urban establishments, roadways, and more. Mapping these features accurately can enable us to monitor and manage various resources effectively.

Highlights

• Deep learning techniques enable accurate water feature mapping
• Two models developed: Unit Model and Encoder/Decoder Model
• Achieved an accuracy score of 99.96%
• Future development includes mapping of other features

FAQ

Q. What is water feature mapping? A. Water feature mapping is the process of extraction and mapping of surface water features such as rivers, streams, lakes, and others.

Q. Why is water feature mapping important? A. Water feature mapping is essential for the management of water resources, disaster management, agricultural monitoring, and much more.

Q. What models were developed for water feature extraction? A. Two models were developed: Unit Model and Encoder/Decoder Model.

Q. Can this model be used to extract other features? A. Yes, in the future, this model can be used to extract other features such as agricultural areas, urban establishments, roadways, and more.