Semantic Segmentation
Overview
This software efficiently performs semantic segmentation on satellite images to classify various features, utilizing both machine learning(ML) and deep learning(DL) techniques.
Features
1. Region of Interest (ROI) Selection and Loading Satellite Imagery:
Selection of Region of Interest. In the advanced tab, users have the option to customize the selection of specific bands for more complex segmentation. The image is then retrieved from the Google Earth Engine's Sentinel-2 database, ensuring high-quality satellite data for analysis.
2. Feature Sampling for ML Models:
Features are sampled from each class to train the machine learning models. Here, the following samples are taken.
3. Machine Learning Model Selection and Thresholds Adjustment:
The selection of the best machine learning model depends on factors like multiclass classification, the presence of background or undefined classes, and overlapping pixel values among classes.
4. Image Segmentation Using Selected Machine Learning Model:
Random Forest Classifier to segement the image.
Tech Stack
Frontend: React, Openstreet Map
Backend: Flask, Python, Scikit-Learn
State Management: React Zustand and Hooks
Styling: Tailwind CSS
Related Links