One of the Ramblings on Convolutional Neural Networks (Introductory Ramblings)_Intefrankly

One of the Ramblings on Convolutional Neural Networks (Introductory Ramblings)

Introductory Ramblings

This article focuses on convolutional neural networks and does the introductory

We know that convolutional neural networks are a very important network architecture in deep learning and have had very significant success in computer vision, why are convolutional neural networks so effective in image processing, what gives it the ability to look at images, we would say, is the samples, is the labeled samples, it's the samples that tell it what's an airplane, what's a bird, what's a hot dog, what's a kielbasa. (Nonsense)

Comprehending images

What is an image? To a computer, an image is a matrix of pixel points. If it is a black and white image, then we only need a matrix to represent it, like this, the matrix on the left represents the image on the right, 30 means black pixel points, 0 means no pixel points, if you follow the annotation of the image you may find that this is a filter representation, but this does not affect the description of the black and white image represented by a matrix

The human brain is a very powerful machine, capable of looking at (capturing) multiple diagrams per second and doing the processing of those diagrams without being aware of it. But that's not the case with machines. The first step for a machine to process an image is to understand, to understand how to represent an image and thus read it.

In simple terms, each image is a series of graph points (pixels) of a particular ordering. If you change the order or color of the pixels, the image changes with it. As an example, store and read an image with the number 4 written on it.

Basically, the machine breaks up the image into a matrix of pixels, storing the color code for each pixel that represents the location. In the representation below, the value 1 is white and 256 is the darkest green (for simplicity, our example is limited to one color).

Once you're done storing image information in this format, the next step is to get the neural network to understand this sorting and pattern.

Convolutional neural network import

Let's talk specifically about convolutional neural networks， Convolutional neural networks are perhaps the most successful case of biology-inspired artificial intelligence， Although convolutional neural networks have also been guided by some other fields， But some of the key designs of neural networks， The principle comes from neuroscience， Two very awesome neurobiologists.， They've made a great discovery.： cat's（ shallow） Patterns of visual neurons for certain specific light（ For example, precisely oriented stripes） It will be very responsive， Almost completely unresponsive to other modes， That tells us one thing.： What if a cat taking a selfie， Some of its visual neurons will be more responsive to its round eyes， Certain neurons will respond more to the stripes on its face， Then the information gradually travels to deeper neurons（ This statement may be inaccurate）， Information will gradually abstract， for example， From the profile， streaks， Abstraction to big eyes， Beautiful floral pattern， In the end it was： wow， A handsome meow.！

The figure below shows that the features extracted by the previous layers are relatively simple, being some color, edge features. The further back the convolutional layers extract the more complex features, being some complex geometric shapes. This is in line with our original design intent for convolutional neural networks, which is to accomplish layer-by-layer feature extraction and abstraction of images through multilayer convolution.

Continuing with the car example below, which is a Tesla model 3, it can be seen that through our convolutional neural network, after each layer

The extracted features are increasingly abstract, we know that if the computer wants to classify this image, there are 10 candidate categories, the computer will calculate a classification for each category of probability, and then classified as the probability of the largest one, we see that the first layer, you can probably know the contours and edges, of course, this can not be used as a basis for classification, the further back, the more abstract the extracted features, to the later, just some complex geometric images, completely unintelligible what the hell, we can not see what the hell, it is normal, the computer knows what the hell, because this is a feature it can understand, using the above human identification car as an analogy, the back of some geometric images that do not know what the hell, for the computer, may be the wheels, intake grille, carriage and so on, finally, these features continue to combine, normalized to probability, and then you can identify the specific category which is

Convolution is a mathematical operation, there is a kind of operation in mathematics called convolution, we can think of a way of feature extraction, convolutional neural network here, there will be multiple filters, each filter represents a specific set of neurons, such as the identification of some stripes, some circle features, these filters are in and our image convolution operation, you can image features layer by layer extraction, combination, abstraction, and finally get the probability of classification, the specific details involved, I will talk in detail in the later articles specific

Recommended>>
1、Multitenant applications using SpringBoot JPA Hibernate and Postgres
2、kafka cluster deployment
3、NginxFPM architecture model profiling and optimization
4、Google publishes paper revealing TPU details 1530x faster than GPUs CPUs
5、Understanding of GC

已推荐到看一看 和朋友分享想法
最多200字，当前共 发送

已发送

确定
分享你的想法...
取消

确定
最多200字，当前共

发送中