Convolutional neural network is suitable for the method of recognizing pictures of various sizes
A technology of convolutional neural network and size, which is applied in the field of recognition of pictures of various sizes and convolutional neural network, can solve the problems of increasing the difficulty of subsequent model training, no effective solution, and reducing the pixels of the picture to be recognized, etc., to achieve fast The effect of effective intelligent recognition
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0052] This embodiment provides a method for convolutional neural network suitable for identifying pictures of various sizes, including the following steps:
[0053] Step 1: A trainable convolutional neural network model architecture for identifying w*h size samples is known; where the w*h size sample is recorded as sample C1, w is the length of sample C1; h is sample C1 width; the known trainable convolutional neural network model architecture includes the following model architecture parameters: in the first convolutional layer after the input layer, the number of feature maps included is n, and the volume used by the first convolutional layer The product kernel size is m*m; where n and m are natural numbers;
[0054] Step 2: Set the original size of the sample to be classified and identified as W*H; wherein, W is the original length of the sample to be classified and identified; H is the original width of the sample to be classified and identified;
[0055] The length and ...
Embodiment 2
[0082] For further understanding of the present invention, a kind of concrete embodiment is introduced below:
[0083] Step 1: If figure 1 As shown, it is a trainable convolutional neural network model architecture for identifying 28*28 size samples;
[0084] exist figure 1 , it can be seen that from the input layer to the output layer, the entire trainable convolutional neural network model architecture has a total of 7 layers. The single-layer perceptron is obtained by rearranging its upper layer, and the influence of this layer is removed. Then remove the input layer and output layer, then, the model architecture parameters of a set of trainable convolutional neural network model architecture are: the number and arrangement of convolutional layers and downsampling layers, that is, after the input layer, there are four layers in total, They are: the first convolutional layer, the first downsampling layer, the second convolutional layer, and the second downsampling layer; i...
Embodiment 3
[0103] This embodiment is basically the same as Embodiment 2, and the only difference is that in this embodiment, a picture of 56*56 is used as the object to be identified, so the first layer is an input layer for inputting a picture of 56*56; The first layer is the image segmentation layer, including four 28*28 sub-images; the third layer is the first convolutional layer, including several feature maps, and a full connection is established between the second layer and the third layer; the fourth layer for figure 1 The second layer of the known trainable convolutional neural network model architecture; the subsequent layers keep the parameters of the known trainable convolutional neural network model architecture unchanged, thus obtaining a new convolutional neural network model architecture.
[0104] Using the newly constructed convolutional neural network model architecture to train 56*56 pictures, the results are as follows Figure 4 shown, from Figure 4 It can be seen t...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More