Notebook
Bearing Faults: FFT Features vs a 1D CNN, With Less Data and More Noise
On the CWRU bearing data both methods score 100% on an unseen motor load, even with four training examples per class. Add noise and the FFT model falls to chance, for a reason that has nothing to do with bearings. Fix that, and the CNN still wins clearly.
Sobel, Scharr or Canny? Edge Detectors Measured Under Noise
Four edge detectors scored against edges known to the pixel, from a clean image to heavy noise. Sobel and Scharr are practically the same detector; what decides the result is smoothing before the gradient and thinning after it, which is exactly what Canny adds.
LMS vs NLMS: Removing Mains Hum from a Real ECG
Two adaptive filters and a plain notch filter clean 60 seconds of a real ECG while the interference suddenly gets four times louder. The step size that works before the jump makes LMS fall apart after it; NLMS keeps working. And the notch filter beats both.
Beyond Labels: How Machines Learn to See With and Without Human Guidance
Explore the fascinating spectrum of learning paradigms in computer vision—from supervised labeling to self-taught machines—and discover how each approach shapes how AI systems understand our visual world.
How Computers Understand Images: A Beginner’s Tour of Image Processing
Explore how computers interpret and manipulate images — from pixels and filters to denoising and contrast enhancement — in this beginner-friendly introduction to digital image processing.
Seeing Signals: How Frequency Shapes the World Around Us
Discover how frequency analysis forms the backbone of modern technology, from image processing to medical imaging, and how understanding frequencies offers a philosophical lens into the nature of reality itself.