Exploratory Work with Satellite Data: Results Assessment

Raya Berova, Thomas Faria, Clément Guillo

13 November 2024

Outline

1️⃣ Introduction

2️⃣ Methodology

3️⃣ Data

4️⃣ Results

5️⃣ Project Structure

6️⃣ Discussion

1️⃣ Introduction

Project Overview

• Satellite imagery exploration for official statistics, initiated through academic challenge (2022)
• Operational implementation in French overseas territories
• Part of European strategy (Warsaw Memorandum)

Applications in Official Statistics

• Cartographic survey optimization:
  • Identification of priority areas for field operations
  • Detection of building changes between surveys
  • Temporal comparison using historical and recent imagery
• Building stock monitoring:
  • Automated change detection
  • Land use evolution
  • Support for population estimates in overseas territories

Focus on Mayotte and French Guiana

• Critical context:
  • Local authorities questioning official statistics
  • Rapid urban development
  • Challenging field conditions
• Satellite data as supporting evidence:
  • Independent source of information
  • Continuous territorial monitoring
  • Transparent methodology

Presentation Outline

1️⃣ Introduction

2️⃣ Methodology

3️⃣ Data

4️⃣ Results

5️⃣ Project Structure

6️⃣ Discussion

2️⃣ Methodology

Semantic Segmentation

Pleiades © CNES_2022, Distribution AIRBUS DS

Training a Segmentation Model

• Model trained to perform automatic segmentation from annotated examples. Requirements:
  • Collection of satellite images
  • Production of annotations (building footprints)
• Model learns to reproduce annotations from images, with the goal to generalize to new images

From Segmentation to Change Detection

Methodology Details

• Model architecture:
  • SegFormer backbone
  • Transformer-based encoder
  • Lightweight decoder

Presentation Outline

1️⃣ Introduction

2️⃣ Methodology

3️⃣ Data

4️⃣ Results

5️⃣ Project Structure

6️⃣ Discussion

3️⃣ Data

Satellite Imagery: Key Features

• Multiple spectral bands beyond RGB (e.g., infrared)
• Georeferenced data: each pixel has specific coordinates
• Main characteristics:
  • Spatial resolution
  • Acquisition frequency
  • Spectral resolution
  • Geographic coverage

Available Data Sources (PLEIADES)

Pléiades (Very High Resolution)

• 0.5m × 0.5m spatial resolution
• 3 bands (RGB)
• Free archives, on-demand acquisition (6-8 months for a department)
• Image size: 1km² (2000 × 2000 pixels)

Available Data Sources (Sentinel-2)

Sentinel-2 (High Resolution)

• 10m × 10m spatial resolution
• 13 spectral bands
• 5-day revisit time, free access
• Image size: 6.25 km² (250 × 250)

Reference Data (BDTOPO)

• Building footprints as vector polygons
• Based on IGN aerial photography
• Used as training data despite temporal misalignment with Pléiades imagery

Presentation Outline

1️⃣ Introduction

2️⃣ Methodology

3️⃣ Data

4️⃣ Results

5️⃣ Project Structure

6️⃣ Discussion

4️⃣ Results

• At this point we have built and trained our model, checked its performance
• Working on notebooks displaying maps… Key questions : how can we share our results ?
• Goals was not to publish an article but rather help colleagues from overseas territory => anticipate carto survey

Interactive Visualization Tool

Access the application: Click here

• Administrative-level monitoring of building changes:
  • Selection by administrative units (municipalities, districts)
  • Quality control of detected changes
• Key features:
  • Side-by-side temporal comparison
  • Statistical indicators by area
  • Interactive visual validation

• Decided to build a tiny webapp … but we are datascientists be indulgent
• We computed building areas from our predictions this we get binary value for each pixel and aggregated by admin level

⚠️ Disclaimer : This is a an early stage of our application. 0 check or data quality process have been done.

• Standard leaflet map with open Street map layer show Pleiades 2023

97611 0332 : nuages in 2022

97610ﾠ0608 : different methodology of blurring for prison

97611 0237 : Our use case : detect building changes in precarious areas

Presentation Outline

1️⃣ Introduction

2️⃣ Methodology

3️⃣ Data

4️⃣ Results

5️⃣ Project Structure

6️⃣ Discussion

5️⃣ Project Structure

Technical Environment

• Cloud-native architecture (SSPCloud platform)
• Modern tech stack: Kubernetes, Docker, MLFlow, React
• Automated workflows for reproducibility
• Continuous deployment practices

Processing Pipeline

1. Need for standardisation of acquisition of data => for now rudimentary. Also need dated images but also in due time without clouds and same methodology => challenge of Pleaides
2. Step for training. Clouds removals/ balancing datasets, labeling and splitting
3. pytorch with pre trained model from HF. Need GPU
4. Build API : predict images, cluster or bbox and compute building area statistics
5. Déploying Geoserver and a web application
6. Colleagues of Insee uses or results to make their statistics

Application Architecture

Geoserver : dynamically stores satellite images

Technical Implementation

• Complex technical environment due to:
  • Non-traditional data processing
  • Large data volumes
  • High computational needs
  • Reproducibility requirements

Challenges & Perspectives

• High maintenance costs due to technical complexity
• Need for specialized skills
• Promising results for cartographic survey support
• Potential improvements identified for each pipeline stage