A Convolutional Neural Network (CNN) is a type of deep learning model that is particularly effective in image classification tasks. Here's a summary of how CNNs work, along with an example:
Key Components:
- Convolutional Layers: These layers scan the input image in small regions, applying filters to extract features such as edges, lines, and curves.
- Pooling Layers: These layers downsample the feature maps, reducing spatial dimensions while retaining important information.
- Flatten Layer: This layer flattens the output of the pooling layer into a 1D vector.
- Fully Connected (FC) Layers: These layers perform classification tasks by processing the flattened input.
Example:
Suppose we want to classify images of dogs and cats as either "dog" or "cat".
Input Image: A color image of size 224x224 pixels
Convolutional Layer 1:
- Filter size: 3x3
- Number of filters: 32
- Activation function: ReLU (Rectified Linear Unit)
- Output shape: 224x224x32
The convolutional layer scans the input image in small regions, applying 32 different filters to extract features such as edges and textures. The output is a feature map with 32 channels.
Max Pooling Layer 1:
- Pool size: 2x2
- Output shape: 112x112x32
The pooling layer downsamples the feature map by taking the maximum value across each 2x2 region, reducing spatial dimensions while retaining important information.
Flatten Layer:
- Input shape: 112x112x32
- Output shape: 1D vector of size 6272 (112x112x32)
The flatten layer transforms the output of the pooling layer into a 1D vector.
Fully Connected Layer 1:
- Number of neurons: 128
- Activation function: ReLU
- Output shape: 1D vector of size 128
The fully connected layer processes the flattened input, performing classification tasks.
Output Layer:
- Number of neurons: 2 (dog or cat)
- Activation function: Softmax
- Output shape: Probability distribution over dog and cat classes
The final output is a probability distribution over the two classes, indicating the likelihood that the image belongs to either a dog or a cat.
This example illustrates how CNNs work for image classification tasks. The architecture consists of convolutional layers for feature extraction, pooling layers for spatial downsampling, flatten layer for input transformation, and fully connected layers for classification.