Table of contents

Full links are relatively cumbersome: Big Fat ManEdit

Edit Many parameters: easy to cause overfitting

Edit Convolution kernel: image feature extraction, derived from the principle of convolution: find the intersection area

The role of convolution

The meaning of convolution

Edit Parameter reduction by convolution kernel

Deep Convolutional Networks

ReLu function: Negative numbers become 0;

ReLu: Removing Bad HabitsEdit

What is pooling: Grasping the main contradictionEdit

Edit average pooling, max pooling

Pooling layer:

What is a Convolutional Neural Network

The full link is relatively bulky: Big Fatty

Many parameters: easy to cause overfitting

Convolution kernel: image feature extraction, derived from the principle of convolution: find the intersection area

Convolution, convolution or convolution (English: Convolution) is a mathematical operation that generates a third function through two functions f and g . Its essence is a special integral transformation , and the characterization functions f and g are reversed. Integral of the product of the overlap function values of and translations over the overlap length

The role of convolution

Feature extraction: Convolution can extract features in the signal through filters, such as edges and textures. These features are very important for image classification and recognition tasks.
Dimensionality reduction: Convolution can reduce the size of the image through the pooling operation, thereby reducing the dimensionality of the data. This is very useful for processing large-scale image and text data.
Denoising: Convolution can remove noise from a signal through a filter. This is very common in the fields of signal processing and image processing and helps to improve the quality of the data.
Image enhancement: convolution can enhance the image through some filters, such as sharpening, smoothing, etc. This helps improve the visuals and quality of the image.

The meaning of convolution

Simulating biological vision: The convolution operation simulates the process of human eyes observing and recognizing images, so convolution is widely used in the field of image processing. It helps us understand how the human visual system works and gives us an efficient way to process images and speech.
Improve algorithm performance: Convolutional neural network (CNN) is currently one of the most important models in deep learning. Its basic structure is the convolution layer. The convolution operation improves the performance of the algorithm in the fields of image recognition, speech recognition and natural language processing. performance. This makes convolutions an essential part of modern machine learning and artificial intelligence.
Data compression: Convolution can reduce the size of data through operations such as dimensionality reduction and filtering, thereby achieving data compression. This is very useful for processing large-scale data, implementing data storage and transmission.

One of the most important applications of the Fourier transform is that the convolution equation can be transformed into a product of two functions to solve. Volume integral is an important member of the integral equation family.

Convolution is an integral operation that can be used to describe the relationship between the input and output of a linear time-invariant system: that is, the output can be obtained by convolution of the input and a function (impulse response function) that characterizes the system characteristics .

Reduce parameters by convolution kernel

Deep Convolutional Networks

The neural network can solve nonlinear problems by adding activation functions, and vice versa is linear regression

ReLu function: Negative numbers become 0;

ReLu: Removing Bad Habits

What is pooling: Grasping the main contradiction

average pooling, max pooling

Pooling layer:

Reduce dimensions, reduce parameters, and avoid overfitting

LeNet: Gradient Learning Convolutional
zfnet reincarnates the role of each layer of a convolutional neural network through visualization.
VGGNET uses stacked small volume sets instead of large convolutions, which can not only increase the judgment of the role function but also reduce the amount of parameters.
googlenet increases the width of the convolutional neural network, adopts a small convolution kernel 1*1, reduces the dimensionality, and reduces the number of parameters; the
resNet residual network solves the degradation problem of the network model, making the neural network deeper.