Improved image classification method based on impulse deep neural network

A pulse depth, neural network technology, applied in biological neural network model, neural architecture, image enhancement and other directions, can solve the problem of back propagation algorithm without biological root, unable to generate time channel, not in line with biological characteristics, etc.

Active Publication Date: 2018-12-11
SHAANXI NORMAL UNIV
View PDF2 Cites 13 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] In recent years, there are many neural networks used for image classification, such as DCNN, SNN, SDNN, etc., but they all have their own shortcomings. Although DCNN has shown good performance in image recognition tasks, the calculation unit of DCNN uses floating point values To represent the activation level of neurons, while organisms communicate by sending electrical pulses, which do not conform to biological characteristics and the backpropagation algorithm of DCNN has no biological roots; the inevitable disadvantage of SNN is that each image requires many pulses and processing time Longer; although SDNN conforms to biological characteristics, it cannot generate different numbers of time channels according to different images

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
  • Improved image classification method based on impulse deep neural network
  • Improved image classification method based on impulse deep neural network
  • Improved image classification method based on impulse deep neural network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0073] The images in this embodiment come from the Caltech 101 data set. The data set contains 101 categories, a total of 8677 images. We select two categories of faces and motorcycles, and select 200 images for each category as a training set, and 198 images for each category as a test set. exist figure 1 , 2 In , an improved spiking deep neural network image classification method consists of the following steps:

[0074] (1) Image preprocessing

[0075] Apply DOG layer and simplified pulse-coupled neural network to preprocess the image, the specific steps are as follows:

[0076] (a) Select an image from the image data set and normalize it to 160 pixels × 250 pixels and grayscale it;

[0077] (b) Generate DOG layer Gaussian filter

[0078]

[0079] In the formula, filt is the Gaussian filter of the DOG layer, s1 and s2 are the standard deviation of the Gaussian filter, and the values ​​are 1 and 2 respectively, and x is a 7×7 matrix, in which the elements of each row...

Embodiment 2

[0118] The images in this embodiment come from the MNIST data set, which contains 10 categories of handwritten digits 0 to 9, a total of 70,000 images, 60,000 images in the training set, and 10,000 images in the test set. We randomly select 500 images of each type from the training set as the training set, a total of 5000 images, and all the test images as the test set. figure 2 Among them, the image classification method of the improved pulse deep neural network of this implementation is made up of the following steps:

[0119] (1) Image preprocessing

[0120] Apply DOG layer and simplified pulse-coupled neural network to preprocess the image, the specific steps are as follows:

[0121] (a) Select an image from the image data set and normalize it to 28 pixels × 28 pixels and grayscale it;

[0122] (b) Generate DOG layer Gaussian filter

[0123]

[0124] In the formula, filt is the Gaussian filter of the DOG layer, s1 and s2 are the standard deviation of the Gaussian fi...

Embodiment 3

[0164] The images in this embodiment come from the Caltech 101 data set. The data set includes 101 categories with a total of 8677 images. We select two categories of airplanes and motorcycles. Each category selects 200 images as a training set, and each category contains 198 images as a test set. . An improved spiking deep neural network image classification method consisting of the following steps:

[0165] (1) Image preprocessing

[0166] Apply DOG layer and simplified pulse-coupled neural network to preprocess the image, the specific steps are as follows:

[0167] (a) Select an image from the image data set and normalize it to 160 pixels × 250 pixels and grayscale it;

[0168] (b) Generate DOG layer Gaussian filter

[0169]

[0170] In the formula, filt is the Gaussian filter of the DOG layer, s1 and s2 are the standard deviation of the Gaussian filter, and the values ​​are 1 and 2 respectively, and x is a 7×7 matrix, in which the elements of each row are 1 to 7 arra...

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

An improved image classification method based on an impulse deep neural network is provided. A DOG layer and a simplified pulse couple neural network are used to preprocess an image, a gray-scale image is generated through the DOG layer to generate a contrast map, and the simplified impulse coupled neural network processes the contrast image generated by DOG layer by parameter adaptive method. According to the different content of the generated contrast image, the larger the pixel value is, the earlier the ignition time is, the impulse image with different number of channels is generated, thatis, the time series impulse image. The improved impulse depth neural network is trained by an STDP unsupervised algorithm. The weight matrix of a convolution layer is updated by STDP weight modification mechanism until the maximum number of iterations of the current convolution layer is reached, and then the training process of the next convolution layer is repeated, and the trained impulse depthneural network is obtained. The method has the advantages of being closer to biological characteristics, being simple and effective, and being suitable for image recognition of handwritten numerals,faces, other objects 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 classifying images. Background technique [0002] Image object classification and detection are two important basic problems in computer vision research. They are the basis of other high-level visual tasks such as image segmentation, object tracking, and behavior analysis. They are also very active research in the fields of computer vision, pattern recognition, and machine learning. direction. Object classification and detection are widely used in many fields, including face recognition, pedestrian detection, intelligent video analysis, pedestrian tracking, etc. in the security field, object recognition in traffic scenes, vehicle counting, retrograde detection, license plate detection and recognition in the traffic field, and Content-based image retrieval in the Internet field, automatic classification of photo albums, etc. [0003] ...

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): G06K9/00G06K9/62G06T5/00G06N3/04
CPCG06T5/009G06V40/172G06V20/10G06V2201/08G06N3/045G06F18/214
Inventor 陈昱莅姚慧婷马苗李兴伟
Owner SHAANXI NORMAL UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap