Skip to main content
Back to module
Lesson 3 of 3

Computer Vision & Language AI

Computer Vision — AI That Sees

How It Works (Layer by Layer)

LAYER 1: Raw pixels (brightness of Red/Green/Blue)

LAYER 2: Edge detection (finds lines and curves)

LAYER 3: Shape detection (combines edges into shapes)

LAYER 4: Part detection (shapes = eyes, ears, nose)

LAYER 5: Object recognition ("This is a cat: 94% confidence")

Ethiopian Vision AI in Action

Researchers are using smartphone cameras + AI to:

  • Identify teff crop diseases from leaf photos
  • Detect coffee cherry ripeness for harvest timing
  • Spot counterfeit Ethiopian coffee in markets

Language AI & The Amharic Challenge

| Language | Speech Recognition Accuracy | Why? |
|:---|:---|:---|
| English | ~95% | Trained on millions of hours of data |
| Mandarin | ~90% | Large investment from Chinese companies |
| Swahili | ~75% | Growing but limited data |
| **Amharic** | **~60-70%** | **Much less training data available** |

Why this matters: Less accurate AI for Amharic means voice assistants don't understand Ethiopian accents well, transcription services are error-prone, and AI translation loses nuance.

The Solution: More Ethiopians creating Amharic digital content = better AI for our language!

Exercise10 points0 attempts

In computer vision, what does the first layer of a neural network typically detect?

Exercise10 points0 attempts

What is the primary reason Amharic speech recognition is less accurate than English?