Computer Vision for Earth Observation: From Manual Digitizing to AI-Powered Analysis

During my graduate school days, I spent countless hours squinting at aerial photos, manually digitizing features and trying to classify false-color infrared images into land cover maps using ERDAS Imagine. If you’ve done similar work, you know the drill — carpal tunnel syndrome looming, endless debates about whether a 3×3 high pass filter was the right choice, and the soul-crushing tedium of pixel-by-pixel classification.

Then came the first time I saw a computer vision model handle these same tasks. I was genuinely in awe. The computer “figured out” all the filters, created all the lines, and handled alignment automatically. It was as if someone had just handed me back weeks of my life. Suddenly, I could focus on my actual research questions instead of the mechanical process of preparing data.

This is the transformative power of computer vision in Earth observation. And if you’ve been watching this space, you’ve likely seen some impressive applications — from automated building footprint extraction to near real-time crop health monitoring. It looks like magic, but it’s not. It’s accessible technology that you can learn to apply to your own work.

You’ve Seen It in Action, Now Learn How to Do It Yourself

If you’re reading this, chances are you spend time working with remote sensing: satellite, aerial photos, drones, lidar… You’ve processed imagery, used QGIS or ArcGIS for analysis, and understand geospatial data structures and formats. You probably know some basic Python — enough to get by, but maybe not enough to confidently implement computer vision models.

That’s exactly where our workshop at FedGeoDay comes in. Getting Started with Computer Vision for Earth Observation is designed for people just like you — professionals who understand the value of remote sensing but want to level up their analysis capabilities with modern AI techniques.

What You’ll Actually Learn (No Buzzwords, Just Skills)

Let’s be clear about what this workshop is: it’s hands-on, practical, and designed to give you skills you can apply immediately. Here’s what we’ll cover:

• Computer Vision Fundamentals: We’ll start with the core concepts you need to understand how these models “see” and interpret image data. Don’t worry, we’ll keep the math to a minimum and focus on intuitive explanations.
• Working with Pre-trained Models: One of the beautiful things about the current state of computer vision is that you don’t need to build everything from scratch. We’ll show you how to leverage existing foundational models to jumpstart your w
• From Images to Insights: We’ll tackle a practical use case that many of you will find immediately applicable: turning multispectral satellite imagery into detailed land cover maps using semantic segmentation.
• Fine-tuning for Your Specific Needs: The real power comes when you can adapt models to your specific requirements. We’ll walk through the process of fine-tuning a model and demonstrate the dramatic improvements this can bring.

All of this will be done with code you can run, modify, and take home with you. No black boxes, no proprietary solutions you can’t afford — just open techniques you can apply to freely available data and models.

Why This Matters for Your Work

Think about how much time you currently spend on preprocessing and basic analysis of remote sensing data. Now imagine dramatically reducing that while simultaneously improving the quality and consistency of your results.

Computer vision can automate the tedious parts of geospatial analysis:

• Identify and extract features without manual digitizing
• Classify land cover consistently across massive datasets
• Detect changes over time with higher sensitivity than manual methods
• Process imagery at scales that would be impossible manually

GePractical Skills, Real Applications

By the end of our workshop, you’ll have:

1. A working environment setup for computer vision with Earth observation data
2. Experience running inference with pre-trained models
3. Practice fine-tuning models for improved performance
4. Code templates you can adapt for your own projects
5. Understanding of best practices and common pitfalls

Most importantly, you’ll have the confidence to continue exploring and applying these techniques after the workshop ends. Our goal isn’t just to show you cool technology — it’s to help you become self-sufficient in using it.

Designed to Meet You Where You are at

Here’s an important note: this workshop assumes you know some basic Python and have some experience with remote sensing. That’s it.

We’re not expecting you to understand backpropagation algorithms or convolutional neural network architectures. We’ll explain what you need to know in plain language and focus on practical implementation rather than theoretical depths.

The goal is to make these powerful techniques accessible to GIS specialists, environmental scientists, urban planners, and anyone else who works with geospatial data — not just machine learning experts.

Join Us on April 23rd in Washington, DC

If you’re ready to transform how you work with Earth observation data, join us at FedGeoDay on April 23rd in Washington, DC. You’ll leave with practical skills, working code, and a solid foundation for continued learning.

Bring your laptop, your curiosity, and your real-world problems. We’ll provide the guidance, the code, and the hands-on experience you need to start applying computer vision to your Earth observation workflows.

Space is limited, and these workshops tend to fill up quickly. We hope to see you there!