Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Convolutional neural network character recognition method running on ARM

A convolutional neural network, character recognition technology, applied in character recognition, neural learning methods, character and pattern recognition, etc., can solve the problem of not being able to meet the recognition accuracy and real-time performance at the same time, not meeting the time and accuracy, and achieving time consumption. Low, speed up the calculation speed, the effect of the efficient calculation method

Pending Publication Date: 2020-12-25
武汉卓目科技有限公司
View PDF0 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, when the recognition scheme based on deep learning is deployed on the hardware device, the two characteristics of recognition accuracy and real-time performance cannot be satisfied at the same time, and the dual requirements of time and accuracy cannot be met.

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
  • Convolutional neural network character recognition method running on ARM
  • Convolutional neural network character recognition method running on ARM
  • Convolutional neural network character recognition method running on ARM

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0053] This embodiment discloses a method for character recognition of a convolutional neural network running on an ARM, such as figure 1 ,include:

[0054] S100. Quantize the convolutional neural network that has been trained as input, and the convolutional neural network with a quantization ratio T, to obtain a quantized network with a weight of 2 to the power of n;

[0055] Specifically, such as figure 2 , the S100 includes:

[0056] S101. Input the trained convolutional neural network, the quantization ratio of each round is set to R, and in some preferred embodiments, R is selected as 10%.

[0057] S102. Determine the quantization weight digit b according to the operation speed limit and quantization accuracy requirements; initialize the quantization mark matrix with the same size as the convolutional neural network, and the quantization mark matrix is ​​a matrix T with all 1s l ;

[0058] Specifically, set the number of digits of the quantization weight to 5, 1 bit ...

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

A convolutional neural network character recognition method running on an ARM comprises the steps that all weight layers of a neural network are quantized into n-th power of 2, and all active layers are quantized into 8-bit signed integers; during calculation of the neural network, all multiplication operations can be replaced by shift operations, the advantage of high efficiency of bit operationsof an ARM processor is fully played, the defect of long time consumption of ARM floating point operations is overcome, and complex neural network calculation can be deployed on embedded equipment with low operational capability and low power consumption.

Description

technical field [0001] The invention belongs to the field of artificial intelligence, in particular to an integer convolutional neural network character recognition method running on an embedded device. Background technique [0002] Generally, character recognition methods are divided into three directions: structural feature recognition, statistical feature recognition and deep learning feature recognition. In structural feature recognition, a recognition judgment tree is formed based on the character structural features and dotted lines of intersecting point features to realize character recognition. The character structure often changes greatly, and even complex situations such as local fractures and adhesions occur. Therefore, the method of structural feature recognition does not have a good result. [0003] In statistical feature recognition, eight-direction gradient features and gradient direction histogram features are mostly used, and two pattern recognition methods...

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
IPC IPC(8): G06N3/04G06N3/063G06N3/08G06K9/20
CPCG06N3/063G06N3/084G06V10/22G06V30/10G06N3/045
Inventor 黄炎杜飞飞
Owner 武汉卓目科技有限公司
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com