News classification
Contact us
- Add: No. 9, North Fourth Ring Road, Haidian District, Beijing. It mainly includes face recognition, living detection, ID card recognition, bank card recognition, business card recognition, license plate recognition, OCR recognition, and intelligent recognition technology.
- Tel: 13146317170 廖经理
- Fax:
- Email: 398017534@qq.com
convolutional neural network
convolutional neural network
In recent years, deep learning has developed rapidly, and has achieved great success in various scenes such as image recognition, speech recognition, and object recognition. For example, AlphaGo beat the world go champion, iPhone X has built-in face recognition and unlocking function and so on. Many AI products have caused great sensation in the world. In this deep learning revolution, the Convolutional Neural Networks (CNN) is the main force to promote all the outbreaks, and has a very important position in the development of artificial intelligence.
.
[question] then what is convolution neural network (CNN)?
1, little white, what is the neural network?
The neural network here, also refers to the Artificial Neural Networks (ANNs), is a mathematical model that mimics the behavioral characteristics of a biological neural network, consisting of a connection between a neuron, a node and a node (synapse), as follows:
002743_sbBk_876354.png
The mathematical model of each neural network unit is abstracted as follows, also called the perceptron, which receives a number of inputs (x1, X2, X3...) to produce an output, which is like the nerve endings that feel the changes in the external environment (external stimuli), and then produce an electrical signal to be transduced to the nerve cells (also called neurons). ).
002750_RNsk_876354.png
A single sensor is a simple model, but in the real world, the actual decision model is much more complex, often a multi-layer network composed of multiple perceptrons. As shown in the following figure, this is also a classic neural network model, consisting of the input layer, the hidden layer and the output layer.
002758_1ChT_876354.png
Artificial neural networks can map arbitrary and complex nonlinear relations, have strong robustness, memory ability, self-learning ability and so on. It is widely used in classification, prediction, pattern recognition and so on.
2, the focus is, what is the convolution neural network?
Convolution neural network has achieved unprecedented accuracy in image recognition and has been widely applied. Next, taking image recognition as an example, we introduce the principle of convolutional neural network.
(1) case
Suppose that a given picture (maybe the letter X or the letter O) can be identified by X as CNN or O. As shown in the following figure, how do we do that?
002814_Ki83_876354.png
(2) image input
If the classical neural network model is used, the entire image needs to be read as the input of the neural network model (that is, the full connection). When the size of the image is larger, the parameters of the connection will become much, which leads to a very large amount of computation.
Our human cognition of the outside world is generally from the local to the global, the perception of the local perception and the gradual cognition of the whole, which is the human cognitive model. The spatial relations in images are also similar. The pixels in the local range are closely related, while the distant pixels are weakly correlated. As a result, it is not necessary for each neuron to perceive the global image, only the local perception is needed, and then the global information is obtained by combining the local information at the higher level. This mode is the important artifact of decreasing the number of parameters in convolution neural network: local receptive field.
002834_i0jn_876354.png
(3) extraction characteristics
If the letter X and the letter O are fixed, the simplest way is to compare the pixels to the pixels, but in real life, there are various forms of changes (such as handwritten word recognition), such as translation, scaling, rotation, micro deformation, etc., as shown in the following figure:
002845_gMdF_876354.png
Our goal is to identify all kinds of X and O through CNN, which involves how to extract features effectively as a key factor in recognition.
Looking back on the "local receptive field" model mentioned earlier, for CNN, it is a small block to compare, to find some rough features (small images) in roughly the same position in the two images. Compared to the traditional one by one, this small block matching method of CNN can be used. It's better to compare the similarity between two images. As follows:
002852_W9U9_876354.png
Taking the letter X as an example, three important features (two crossover lines and one diagonal line) can be extracted, as shown in the following figure:
002859_iMEv_876354.png
If the pixel value "1" represents white, and the pixel value "-1" represents black, the three important characteristics of the letter X are as follows:
002907_aiwg_876354.png
How do these characteristics match computation? Don't talk to me about pixel matching. Sweat!
(4) convolution (Convolution)
Then we must invite today's important guests: convolution. Then what is convolution? It's not urgent. Let's go slowly down here.
When a new graph is given, CNN does not know exactly what parts of the original map are to be matched, so it will try every possible location in the original diagram, equivalent to turning the feature into a filter. The process used to match is called convolution operation, which is also the origin of convolutional neural network.
.
[question] then what is convolution neural network (CNN)?
1, little white, what is the neural network?
The neural network here, also refers to the Artificial Neural Networks (ANNs), is a mathematical model that mimics the behavioral characteristics of a biological neural network, consisting of a connection between a neuron, a node and a node (synapse), as follows:
002743_sbBk_876354.png
The mathematical model of each neural network unit is abstracted as follows, also called the perceptron, which receives a number of inputs (x1, X2, X3...) to produce an output, which is like the nerve endings that feel the changes in the external environment (external stimuli), and then produce an electrical signal to be transduced to the nerve cells (also called neurons). ).
002750_RNsk_876354.png
A single sensor is a simple model, but in the real world, the actual decision model is much more complex, often a multi-layer network composed of multiple perceptrons. As shown in the following figure, this is also a classic neural network model, consisting of the input layer, the hidden layer and the output layer.
002758_1ChT_876354.png
Artificial neural networks can map arbitrary and complex nonlinear relations, have strong robustness, memory ability, self-learning ability and so on. It is widely used in classification, prediction, pattern recognition and so on.
2, the focus is, what is the convolution neural network?
Convolution neural network has achieved unprecedented accuracy in image recognition and has been widely applied. Next, taking image recognition as an example, we introduce the principle of convolutional neural network.
(1) case
Suppose that a given picture (maybe the letter X or the letter O) can be identified by X as CNN or O. As shown in the following figure, how do we do that?
002814_Ki83_876354.png
(2) image input
If the classical neural network model is used, the entire image needs to be read as the input of the neural network model (that is, the full connection). When the size of the image is larger, the parameters of the connection will become much, which leads to a very large amount of computation.
Our human cognition of the outside world is generally from the local to the global, the perception of the local perception and the gradual cognition of the whole, which is the human cognitive model. The spatial relations in images are also similar. The pixels in the local range are closely related, while the distant pixels are weakly correlated. As a result, it is not necessary for each neuron to perceive the global image, only the local perception is needed, and then the global information is obtained by combining the local information at the higher level. This mode is the important artifact of decreasing the number of parameters in convolution neural network: local receptive field.
002834_i0jn_876354.png
(3) extraction characteristics
If the letter X and the letter O are fixed, the simplest way is to compare the pixels to the pixels, but in real life, there are various forms of changes (such as handwritten word recognition), such as translation, scaling, rotation, micro deformation, etc., as shown in the following figure:
002845_gMdF_876354.png
Our goal is to identify all kinds of X and O through CNN, which involves how to extract features effectively as a key factor in recognition.
Looking back on the "local receptive field" model mentioned earlier, for CNN, it is a small block to compare, to find some rough features (small images) in roughly the same position in the two images. Compared to the traditional one by one, this small block matching method of CNN can be used. It's better to compare the similarity between two images. As follows:
002852_W9U9_876354.png
Taking the letter X as an example, three important features (two crossover lines and one diagonal line) can be extracted, as shown in the following figure:
002859_iMEv_876354.png
If the pixel value "1" represents white, and the pixel value "-1" represents black, the three important characteristics of the letter X are as follows:
002907_aiwg_876354.png
How do these characteristics match computation? Don't talk to me about pixel matching. Sweat!
(4) convolution (Convolution)
Then we must invite today's important guests: convolution. Then what is convolution? It's not urgent. Let's go slowly down here.
When a new graph is given, CNN does not know exactly what parts of the original map are to be matched, so it will try every possible location in the original diagram, equivalent to turning the feature into a filter. The process used to match is called convolution operation, which is also the origin of convolutional neural network.