Introduction to Convolutional Neural Networks
Learn how convolutional neural networks detect visual patterns and build image predictions.
Introduction to Convolutional Neural Networks explained visually
Follow a bird image through convolution, activation, pooling, and classification.
What you will learn
Follow filters, pooling, flattening, and classification through complete CNN examples.
IMAGE INPUT
A CNN examines small image regions instead of connecting every pixel to every neuron.
LOCAL FILTERS
Learnable filters respond to useful visual patterns such as edges, curves, and textures.
ACTIVATION
Activation and pooling shape the feature maps while reducing unnecessary spatial detail.
POOLING
Deeper layers combine learned features and produce the final visual prediction.
Quick check: test your understanding
What does a convolutional filter search for?
A learned local visual pattern.
Why use pooling?
To reduce spatial dimensions while preserving important information.
What creates the final prediction?
Deeper features are combined by the network's final layers.
Accessible referenceLesson 41 transcript313 words · Matches the published narration and captionsOpen transcript
Imagine that you want a computer to recognize a bird in a photograph. A standard fully connected neural network can process the image, but it treats every pixel as a separate input connected to many neurons. For a large image, this creates an enormous number of connections and parameters.
A Convolutional Neural Network, commonly called a CNN, processes images in a more practical way. A CNN examines small regions of an image and searches for visual patterns. These patterns can include edges, corners, textures, and shapes. The first major component is the convolutional layer.
A convolutional layer uses small learnable matrices called filters or kernels. Each filter moves across the image and responds strongly when it finds a particular pattern. One filter may learn to recognize vertical edges. Another may react to curves or changes in texture. After convolution, an activation function such as ReLU introduces nonlinearity.
This allows the network to represent relationships that a simple linear calculation cannot capture. A CNN may also contain pooling layers. Pooling reduces the width and height of feature maps while preserving important information. This lowers computational cost and makes the network less sensitive to small movements in an object.
Near the end of the network, fully connected layers combine the extracted features and produce a final prediction. For example, early layers may detect edges in the photograph. Deeper layers may combine those edges into a beak, wings, and feathers. The final layer uses those features to predict that the image contains a bird.
CNNs support many computer vision tasks. They can classify images, detect objects, segment individual pixels, and analyze video. The central idea is simple. A CNN learns small visual patterns first, combines them into larger patterns, and uses those patterns to understand the complete image. In the next lesson, we will examine spatial structure and explain why the position of visual features matters.