AI Trend Research

← AI glossary

AI glossary · Basic · Computer vision

CNN

A CNN (convolutional neural network) is an artificial neural network that finds patterns in grid-shaped data such as images by sliding a small window (a filter) across it step by step. It has been widely used as the foundation of technology that recognizes objects in photos.

About this term
Level
Basic · Technical terms you often meet in AI news
Added
2026-10-08
In plain words It is like scanning a large picture square by square with a magnifying glass, marking 'there's a line here' or 'there's a circle here'. The difference is that a real CNN is not told what to look for; it learns that on its own from data.

How it works

The core of a CNN is the convolutional layer. A small filter slides across the image one step at a time and scores whether a particular pattern (a horizontal line, a corner and so on) appears at each position. Next, a pooling step summarizes this result coarsely to shrink its size, which helps the network recognize an object similarly even if its position shifts slightly. When many such layers are stacked, early layers recognize lines and edges, and later layers recognize more complex shapes such as eyes or wheels. A final classification layer uses the collected features to produce a result such as 'cat' or 'car'.

How a CNN recognizes an imageInput image → Convolutional layer; Convolutional layer → Pooling; Pooling → Deeper layers; Deeper layers → Classification layer; Classification layer → ResultInput imagePixel gridConvolutionallayerFilters find lines andedgesPoolingSummarize by shrinkingDeeper layersShapes like eyes andwheelsClassificationlayerCombine features todecideResulte.g. cat
How a CNN recognizes an image Moving left to right through the layers, features build up from simple lines to complex shapes, and a classification result comes out at the end. 'Cat' is an example.

Why it matters

Connecting every pixel of an image all at once would require far too many parameters. Because a CNN reuses the same filter across the entire image, it learns efficiently with far fewer parameters. This drove major progress in areas such as photo classification, face recognition, medical image reading and camera perception for self-driving cars. Much of how deep learning became known to the public came from CNN successes in image recognition.

Things to know

Transformer-based models are now widely used for images as well, but CNNs are still common in on-device AI such as smartphones because they are lightweight and fast. Many models also combine the two approaches. CNNs capture relationships between nearby pixels well, but need many stacked layers to grasp relationships between distant parts of an image. They can also respond in strange ways to images that differ greatly from their training data.

Example

CNN-based models are commonly used when a smartphone photo gallery automatically sorts pictures into categories like 'food', 'pets' and 'landscapes', or when a camera app finds faces to set focus. Factory defect inspection systems that photograph products to spot scratches are another typical use.

Related terms

Artificial neural network Deep learning Transformer Activation function Backpropagation On-device AI