Unlock instant, AI-driven research and patent intelligence for your innovation.

Construction method of image classified CNN (Convolutional Neural Network) structure

A convolutional neural network and convolutional neural technology, applied in the field of convolutional neural network structure, can solve the problems of increasing the number of network parameters and increasing the complexity of network calculations, achieving reduced complexity, improved classification performance, and simple construction methods Effect

Active Publication Date: 2017-06-20
SHAANXI NORMAL UNIV
View PDF2 Cites 29 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] The main technical problem in the above-mentioned image classification method based on neural networks is that in the process of network structure design, the number and size of filters, pooling methods, and activation functions are often determined by experience, and inappropriate network structures will greatly affect Increase the number of network parameters, leading to an increase in the complexity of network calculations

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Construction method of image classified CNN (Convolutional Neural Network) structure
  • Construction method of image classified CNN (Convolutional Neural Network) structure
  • Construction method of image classified CNN (Convolutional Neural Network) structure

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0054] The image of this embodiment comes from the SVHN (The Street View House Numbers Dataset) dataset that Google extracts from the street view house number image in the real world. The 73257 images in this dataset are used in this embodiment as In the training set, 26032 images are used as the test set, and the training set and the test set do not overlap.

[0055] exist figure 1 Among them, the construction method of the image classification convolutional neural network structure of the present embodiment is composed of the steps of constructing the convolutional neural network structure, convolutional neural network training and testing, and the steps of constructing the convolutional neural network structure are as follows:

[0056] (1) Obtain training sample images and test sample images and preprocess them

[0057] (a) Select 73257 training sample images and 26032 testing sample images from the image dataset.

[0058] (b) Preprocessing 73257 training sample images

...

Embodiment 2

[0094] The images in this example come from the MNIST dataset consisting of handwritten digits. In this embodiment, 60,000 digital images in the data set are used as a training set, and 10,000 digital images are used as a test set, and the training set and the test set do not overlap.

[0095] The construction method of the image classification convolutional neural network structure of the present embodiment is made up of construction convolutional neural network structure, convolutional neural network training and testing steps, and the steps of constructing convolutional neural network structure are as follows:

[0096] (1) Obtain training sample images and test sample images

[0097] (a) Select 60,000 training sample images and 10,000 testing sample images from the image dataset.

[0098] (b) Preprocessing 60,000 training sample images

[0099] The steps for pretreatment are the same as in Example 1.

[0100] (c) Preprocess 10,000 test sample images

[0101] The preproc...

Embodiment 3

[0120] The images in this embodiment come from an ASL (American Sign Language, ASL) data set composed of gesture images. In this embodiment, 50,400 gesture images in the data set are used as a training set, and 6,000 gesture images are used as a test set, and the training set and the test set do not overlap.

[0121] The construction method of the image classification convolutional neural network structure of the present embodiment is made up of construction convolutional neural network structure, convolutional neural network training and testing steps, and the steps of constructing convolutional neural network structure are as follows:

[0122] (1) Obtain training sample images and test sample images and preprocess them

[0123] (a) Select 50400 training sample images and 6000 testing sample images from the image dataset.

[0124] (b) Preprocessing 50400 training sample images

[0125] Preprocessing the training sample images is the same as that in Embodiment 1.

[0126] (...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention relates to a construction method of image classified CNN structure. The method comprises a step of constructing the CNN structure and a step of training and testing the CNN. In the step of constructing the CNN structure, the image classified CNN structure is obtained by obtaining and preprocessing training sample images and test sample images, setting an initial structure of the CNN, introducing an activation function, determining a pooling method, determining a filter and determining the number of convolution layers. According to the step of training and testing the CNN, a K-Means clustering method is used to obtain a filter of the convolution layer with 20 characteristic graphs, a network weight matrix is updated to a maximal training frequency via forward and backward spreading, a trained CNN is obtained, the test sample images are tested, and a verified image classified CNN structure is obtained. The construction method has the advantages of being simple and effective, and is suitable for image classification of doorplate numbers, handwritten numbers, zip codes and the like.

Description

technical field [0001] The invention belongs to the technical field of image processing and pattern recognition, and in particular relates to a convolutional neural network structure suitable for image classification. Background technique [0002] Deep learning is a new field in machine learning research. Its motivation is to establish and simulate the neural network of human brain for analysis and learning. It interprets data, such as images, sounds, and texts, by imitating the mechanism of the human brain. Its core is to learn more useful features by building a machine learning model with multiple hidden layers and a large amount of training data, thereby ultimately improving classification or forecast accuracy. [0003] Existing representative networks such as LeNet-5, AlexNet, GoogLeNet, and ResNet are all based on convolutional neural network structures. Among them, LeNet-5 consists of 7 layers, its 1st, 3rd, and 5th layers are convolutional layers, and 2nd and 4th la...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/62G06N3/08
CPCG06N3/08G06N3/084G06F18/23213
Inventor 马苗刘琳武杰陈昱莅裴炤
Owner SHAANXI NORMAL UNIV