LeNet-5 is a classic convolutional neural network (CNN) architecture, which was proposed in the late 1980s. It played an important role in the early development of CNNs and laid the foundation for the subsequent development of deep learning.

Overview

  • Layer Structure: LeNet-5 consists of two convolutional layers, two pooling layers, and three fully connected layers.
  • Input Size: The input image size is 32x32 pixels.
  • Output: The output layer has 10 neurons, corresponding to the 10 possible digit classes (0-9).

Architecture Details

Convolutional Layers

  • Conv1: 6 filters, 5x5 size, stride 1, padding 2.
  • Conv2: 16 filters, 5x5 size, stride 1, padding 2.

Pooling Layers

  • Pool1: 2x2 pooling with stride 2.
  • Pool2: 2x2 pooling with stride 2.

Fully Connected Layers

  • FC1: 120 neurons, using the sigmoid activation function.
  • FC2: 84 neurons, using the sigmoid activation function.
  • FC3: 10 neurons, using the softmax activation function.

Example

Here is an example of the LeNet-5 architecture:

[Input] -> [Conv1] -> [Pool1] -> [Conv2] -> [Pool2] -> [FC1] -> [FC2] -> [FC3] -> [Output]

Image

LeNet-5 Architecture

For more information about CNNs, please refer to our CNN Tutorial.