One, the principle of vision
1. In 1981, David Huber and Tostan Weizer; experiment on cats (1, simple lines 2, complex movements)
2. Kunihiko Fukushima (neurocognitive model) Hikari ---V1----V2----V3
Pixels-----edge direction-----object outline and details-----information is integrated together to judge the object
3. Yang Likun: Convolutional Neural Network CNN (Image Recognition)
2. Convolutional neural network (mathematical method to extract the features of the image)
| 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| 0 | 1 | 0 | 0 | 0 | 1 | 0 |
| 0 | 0 | 1 | 0 | 1 | 0 | 0 |
| 0 | 0 | 0 | 1 | 0 | 0 | 0 |
| 0 | 0 | 1 | 0 | 1 | 0 | 0 |
| 0 | 1 | 0 | 0 | 0 | 1 | 0 |
| 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| 1 | 0 | 0 |
| 0 | 1 | 0 |
| 0 | 0 | 1 |
| 2 | 0 | 1 | 0 | 1 |
| 0 | 3 | 0 | 1 | 0 |
| 1 | 0 | 3 | 0 | 1 |
| 0 | 1 | 0 | 3 | 0 |
| 1 | 0 | 1 | 0 | 2 |
Note: Different convolution kernels produce different features. It is important to choose a convolution kernel (above example: the convolution kernel on the lower and upper right side convolves the original image. The larger the number, the more obvious the feature is. 333 The feature is obvious)
Three, pooling, activation
Feature map
| 2 | 0 | 1 | 0 | 1 |
| 0 | 3 | 0 | 1 | 0 |
| 1 | 0 | 3 | 0 | 1 |
| 0 | 1 | 0 | 3 | 0 |
| 1 | 0 | 1 | 0 | 2 |
| 3 | 1 | 1 |
| 1 | 2 | 1 |
| 1 | 1 | 2 |
Choose the largest value in the color block and fill it in the 3x3 form, which is pooling
Activate with the sigmode function to activate
| 0.95 | 0.73 | 0.73 |
| 0.73 | 0.95 | 0.73 |
| 0.73 | 0.73 | 0.88 |
Activate to a value in the range (0-1).
Four, algorithm, data, computing power
Introduction to computing power: computing power is the efficiency of computing (chip)
1. CPU: versatility is better, cpu is the central processing unit, features, poor parallel capability, and good serial capability.
(1. Parallel: Calculate multiple additions and subtractions at the same time. 2. Serial: Only calculate one by one)
2. GPU: (computer image projected on the screen) represents the company's Nvidia, which has relatively strong specificity, poor serial capability, good parallel capability, and calculates multiple addition and subtraction at the same time.
3. NPU: Neural network processor, specificity and parallelism are stronger than GPU; Huawei's Shengteng 910NPU processor. (Currently the fastest artificial intelligence processor)
