Depthwise fast convolution system based on image processing, and image recognition method

A fast convolution and image processing technology, applied in image data processing, image enhancement, instruments, etc., can solve problems such as consumption and multi-resources, reduce computing space, high utilization rate, and avoid repeated reading of image storage units. effect of the process

Pending Publication Date: 2020-06-09
中科南京人工智能创新研究院
View PDF2 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0006] 2) After the first reading calculation, the data results need to be stored, which will consume more resources

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
  • Depthwise fast convolution system based on image processing, and image recognition method
  • Depthwise fast convolution system based on image processing, and image recognition method
  • Depthwise fast convolution system based on image processing, and image recognition method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0041] In this embodiment, a Depthwise fast convolution system based on image processing includes an image pixel reading unit, a DW convolution calculation unit, and a detection and comparison unit, and realizes image convolution by using rows as data units and performing operations with convolution kernels. area, reducing repeated reading of the image storage unit;

[0042] The image pixel reading unit marks each group of pixels in the image in the three dimensions of H, W and D, stores the image to be processed in the storage unit, and first considers the D dimension according to the calculation parallelism, then the H dimension, and finally is the W dimension;

[0043] DW convolution calculation unit, such as figure 1 As shown, after the image data is read by row, each row of image data must be multiplied and accumulated with three rows of convolution kernels until the calculation of all image data rows is completed, and finally the pixel is multiplied by the corresponding...

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 discloses a Depthwise fast convolution system based on image processing. The Depthwise fast convolution system comprises an image pixel reading unit and a DW convolution calculation unit. The image pixel reading unit can store an image to be processed into the storage unit in a multi-dimensional form; and the DW convolution calculation unit reads the image data in rows and then performs multiply-accumulate calculation with the convolution kernels of the three rows to obtain calculation results of all pixel points. According to the system, a rapid algorithm of image convolution isrealized, the process of repeatedly reading the image storage unit can be reduced, the calculation space used by convolution is reduced, and the result of image convolution can be rapidly and accurately obtained.

Description

technical field [0001] The invention relates to an image convolution technology, in particular to a Depthwise fast convolution system based on image processing. Background technique [0002] With the continuous advancement of electronic technology, the usage rate of images and videos in mainstream media has gradually increased, and has steadily occupied a dominant position, replacing people's demand for text. [0003] In the field of machine learning or neural network computing, the processing of images and videos requires further processing such as storing the image to be processed in a storage unit in a certain format, and then performing convolution on the data by repeatedly reading image pixels. calculate. [0004] The commonly used way to realize image convolution is to store image data in three dimensions, take data in fixed windows along two dimensions, and further complete the acquisition of the product value of internal data. Although this method of reading data i...

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): G06T5/50G06N3/04
CPCG06T5/50G06N3/045
Inventor 李钢周欢欢张旸
Owner 中科南京人工智能创新研究院
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