Convolutional Neural Network – Simple Explanation



Convolutional Neural Network: Definition

Convolutional Neural Network or CNN is an artificial neural network. The word ‘Convolution’ stands for mathematical operation, which involves merging two sets of data to form a third one.

Neural Network is a series of algorithms that reproduces the operation of a human brain. The aim of a Neural Network is to recognize relationships between huge amounts of data.

CNN analyzes images and data. Mainly to pick out and detect patterns and make sense of them.

Convolutional Neural Network Diagram

How does CNN operate?

Convolutional layers are hidden layers. They receive input, transform it and produce an output to the next layer. For example, image patterns are detected using Filters in convolutional layers.

Convolutional Neural Network: Image Patterns
CNN: Image Patterns

Filter is a small matrix (3×3 pixels). Filters detect shapes, objects, edges and many more, which are patterns by definition. Filters will convolve / slide through each block of 3×3 pixels and create new representation of an image.



CNN: Filters

In this case, Filter 1 detects the upper edge of an image, indicated by ‘bright’ (Yellow Arrow) which corresponds to black. Filter 2 detects lower edge of an image, indicated by ‘bright’ (Yellow Arrow) which corresponds to black too.

CNN: 3 Applications

Medicine. CNN can analyze an image of a part of human body, where cancer may be present. Additionally, it can detect broken bones in X-Ray Images.

Vehicles. CNN can be used in driverless cars, for robots which should reproduce human behavior. CNN can analyze input images of the road and detect key ‘patterns’ such as obstacles or pedestrians.

Technology. CNN is used for facial recognition. Hence it is useful in various fields such as photo editing. Moreover, it can recognize different objects, thus enabling search by image in google and many more.

I hope you found this article useful!

Comment your opinion on CNN below and any additional points to include!



Link to the main publication