Article Preview
Top2. Convolution Neural Network (Cnn)
The deep learning model is a multi-layer mode constructed by integrating the convolution layer and the hidden layer. The common BP neurons are shown in Figure 1, and the corresponding functions are as follows (Aiqin et al., 2014):
(1)In formula (1), f is the activation function, usually is a nonlinear function, such as sigmoid, ReLu, etc., and the neural network model is composed of multiple neurons. Figure 2 shows a three-layer BP neural network structure. Images can be represented in pixels, each pixel can represent one dimension, and the image of 100 × 100 is represented as 10000 dimensions correspondingly, and each dimension is represented as vector. If the dimensions of the input image set and the hidden image set were the same, the corresponding order of magnitude of the vector would be 100002. The order of magnitude is too large, and there is serious redundancy.
Figure 2. Three layer BP neural network structure
We use CNN to realize local visual field perception and the same weight in the same channel is used to reduce the order of magnitude of vector. The local visual field perception is based on the observation features of the human eye from the local to the overall. The hidden layer is only associated with the neuron set of the previous layer in the multi-layer neural network structure, so that the vector order of magnitude of the product will be greatly reduced, as shown in Figure 3 (Yang et al., 2014).
Figure 3.
Local visual field perception