Automated Building Detection via Satellite Imagery
Budget: €250 – €750 EUR
We are seeking an experienced Python developer to build a robust, automated software for detecting building footprints from satellite imagery. The goal is to create a pipeline that takes a user's Area of Interest (AOI) and a reference date, and in response, generates a detailed GeoJSON file of all detected building footprints.
This solution will leverage super-resolved Sentinel-2 imagery. The workflow will involve processing a cloud-free image composite, applying a super-resolution model, and then running a deep learning model for the core detection task.
A key part of this project is the integration of existing Python software components (which will be provided) for cloud-free image generation and super-resolution. The selected freelancer will be responsible for building the main application, integrating these modules, and developing the new deep learning component for building detection.
2. Understanding of the Project
I have thoroughly reviewed the project requirements. The operational workflow will be as follows:
User Inputs:
Area of Interest (AOI) in GeoJSON, KML, or WKT format.
A reference date.
Band selection (RGB or all bands) for super-resolution.
Automated Workflow:
Cloud-Free Composite: The software will search for Sentinel-2 images prior to the reference date, mosaic them as needed to cover the AOI, and generate a single, cloud-free standard-resolution image.
Super-Resolution (SR): The cloud-free image will be processed to create a super-resolution version. For large AOIs, this will be handled in 10km x 10km tiles to ensure scalability. The final SR image(s) will be saved in Cloud-Optimized GeoTIFF (COG) format.
Building Detection: A deep learning model will process the SR image(s) to detect building footprints.
Output Generation: The results will be vectorized and saved as a single GeoJSON file, with each building feature including an ID, center position (lat/lon), polygon geometry, area (e.g., in sq. meters), and a confidence score for the detection.
This solution will leverage super-resolved Sentinel-2 imagery. The workflow will involve processing a cloud-free image composite, applying a super-resolution model, and then running a deep learning model for the core detection task.
A key part of this project is the integration of existing Python software components (which will be provided) for cloud-free image generation and super-resolution. The selected freelancer will be responsible for building the main application, integrating these modules, and developing the new deep learning component for building detection.
2. Understanding of the Project
I have thoroughly reviewed the project requirements. The operational workflow will be as follows:
User Inputs:
Area of Interest (AOI) in GeoJSON, KML, or WKT format.
A reference date.
Band selection (RGB or all bands) for super-resolution.
Automated Workflow:
Cloud-Free Composite: The software will search for Sentinel-2 images prior to the reference date, mosaic them as needed to cover the AOI, and generate a single, cloud-free standard-resolution image.
Super-Resolution (SR): The cloud-free image will be processed to create a super-resolution version. For large AOIs, this will be handled in 10km x 10km tiles to ensure scalability. The final SR image(s) will be saved in Cloud-Optimized GeoTIFF (COG) format.
Building Detection: A deep learning model will process the SR image(s) to detect building footprints.
Output Generation: The results will be vectorized and saved as a single GeoJSON file, with each building feature including an ID, center position (lat/lon), polygon geometry, area (e.g., in sq. meters), and a confidence score for the detection.
Related categories:
JavaScript
Python
Django
Software Architecture
Machine Learning (ML)
Data Science
Computer Vision
Deep Learning