Convolutional neural recognition system and method based on ARM and FPGA

A convolutional neural and identification system technology, applied in neural architecture, biological neural network model, architecture with a single central processing unit, etc., can solve the problems of poor flexibility, long design cycle, difficult popularization, etc., to provide flexibility , the effect of fast operation and high flexibility

Pending Publication Date: 2020-10-30
上海仪电(集团)有限公司中央研究院 +1
View PDF4 Cites 5 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

At present, the chip deployment methods of convolutional neural network mainly include: (1) The calculation method of GPU+CPU combination has become the preferred solution for machine learning. Among them, CPU is more suitable for human-computer interaction and process control due to its versatility; Compared with the CPU block, the high price and large power consumption make it difficult to be used in low-end occasions; (2) Although ASIC is the most cost-effective convolutional neural network computing chip, but also because of the long design cycle and high cost, There are also many problems with poor flexibility; (3) FPGA programmable chips have high requirements for implanting servers, programming environments, and developer capabilities, and lack general potential, making it difficult to popularize

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 recognition system and method based on ARM and FPGA
  • Convolutional neural recognition system and method based on ARM and FPGA
  • Convolutional neural recognition system and method based on ARM and FPGA

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0041] like figure 1 As shown, the present embodiment provides a convolutional neural recognition system based on ARM and FPGA, the system includes a convolutional neural network parameter training module, an ARM chip and an FPGA convolutional neural network hardware module, and the convolutional neural network parameter training module uses It is used to obtain the model parameters of the convolutional neural network; the ARM chip is used to receive and forward the model parameters, and transmit the recognition result to the display device; the FPGA convolutional neural network hardware module is used to realize the All convolutional neural networks with parameters from the input layer, convolutional layer, pooling layer, activation layer to the fully connected layer perform forward prediction operations on the real-time input image stream, realize image recognition and classification, and output recognition results.

[0042] In this embodiment, the convolutional neural netwo...

Embodiment 2

[0047] like figure 2 As shown, the present embodiment provides a convolutional neural recognition method based on ARM and FPGA, the method comprising the following steps:

[0048] (1) Collect the picture library to be trained, use caffe to train the picture library on the PC, generate model parameter files, extract parameters through tools, download them to ARM through USB, and ARM stores the data in FALSH.

[0049] (2) After each power-on and start-up, ARM reads the parameters in the FLASH, and configures the data into the FPGA through the bus interface with the FPGA, and the FPGA stores the data in the Distributed RAM of each module in turn.

[0050] (3) After power-on initialization of CAMERA and parameter configuration, start to read the image stream S0 of CAMERA.

[0051] (4) CAMERA outputs a 640x480 resolution image stream S0, which first forms an image window through the windows module, and after convolution operation with a fixed convolution kernel C0, outputs an ima...

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 convolutional neural recognition system and method based on ARM and FPGA, and the system comprises: a convolutional neural network parameter training module which is used for obtaining the model parameters of a convolutional neural network; an ARM chip which is used for receiving and forwarding the model parameters and transmitting an identification result to display equipment; and an FPGA convolutional neural network hardware module which is used for achieving all convolutional neural networks with the model parameters from an input layer, a convolutional layer, a pooling layer, an activation layer to a full connection layer in a hardware circuit mode, performing forward prediction operation on the image flow input in real time, achieving image identification classification and outputting an identification result. Compared with the prior art, the system has the advantages of low cost, low power consumption, high operation speed and the like.

Description

technical field [0001] The invention relates to a method for realizing a convolutional neural recognition system, in particular to a convolutional neural recognition system and method based on ARM and FPGA. Background technique [0002] From the convolutional neural network ImageNet winning the championship with a 16% error rate, to AlphaGo defeating Lee Sedol, to the current intelligent multilingual translation, weather forecast, speech recognition, intelligent security, artificial intelligence, driverless and other systems, with the development of algorithms Continuously evolving and improving, artificial intelligence has been applied to every corner of our lives. As the engine of the third wave of artificial intelligence, deep learning is usually attributed to three points: big data, supercomputing power and new mathematical methods. [0003] The data needed for sufficient training of the deep learning model should benefit from the rapid development of the Internet and t...

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): G06N3/04G06N3/063G06F15/78
CPCG06N3/063G06F15/7817G06N3/045Y02D10/00
Inventor 左佳李少华张青野赵中瑞方逸洲
Owner 上海仪电(集团)有限公司中央研究院
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