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

Image processing method and device based on convolutional neural network model

A convolutional neural network and neural network technology, applied in biological neural network models, image data processing, neural architecture, etc., can solve problems such as large energy consumption

Active Publication Date: 2021-09-21
HUAWEI TECH CO LTD
View PDF7 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

From the perspective of the required storage space, the storage and transmission of model parameters also consume a lot of energy

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
  • Image processing method and device based on convolutional neural network model
  • Image processing method and device based on convolutional neural network model
  • Image processing method and device based on convolutional neural network model

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0038] The technical solutions in the present application will be described below in conjunction with the drawings.

[0039]For ease of understanding, the first neural network is described in detail. Neural networks generally include a plurality of layers of neural networks, each neural network may be implemented in different layers of operation or operations. Common layer comprises a convolution neural network layer (convolutionlayer), pooled layer (pooling layer) and the fully connected layers (full-connection layer) and the like.

[0040] figure 1 It is a convolutional neural network (convolutional neural networks, CNN) basic framework FIG. See figure 1 , Convolutional neural network comprises a layer of a convolution, and the cell layer fully connected layer. Wherein the plurality of layers and a plurality of convolution cell layers are alternately arranged, after convolution convolution layer may be a layer, layer may be pooled.

[0041] Convolution input matrix layer is mai...

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 present application provides an image processing method and device based on a convolutional neural network model. The method includes: obtaining a first weight parameter set corresponding to a neural network layer, the first weight parameter set including N1 first weight parameters, wherein N1 is an integer greater than or equal to 1; calculating the N1 first weight parameters and the first weight parameters respectively The ratio of a value m to get N1 second weight parameters, where |W max |≤m≤2|W max |, W max is the weight parameter with the largest absolute value in the first weight parameter set; N1 second weight parameters are respectively quantized into the sum of at least two Q powers of 2 to obtain N1 third weight parameters, where Q≤0, and Q is an integer; acquire the image to be processed; process the image to be processed according to N1 third weight parameters to obtain an output image. The present application can reduce errors caused by weight quantization, thereby reducing precision loss.

Description

Technical field [0001] The present application relates to the field of image processing, and more particularly to an image processing method and apparatus based on convolutional neural network model. Background technique [0002] In recent years, neural networks, especially convolutional neural networks, have achieved huge success on image processing, image recognition class applications. A typical convolutional neural network generally makes a plurality of convolutional layers, all connect layers, and the like, from the calculation angle, multiplication is the main bottleneck. From the desired storage space, model parameters storage and transmission also need to consume a lot of energy. Many researchers have studied the method of compressing and accelerating the neural network, that is, while the model storage space is reduced, the calculation amount (ie, the number of multiplication) can also be significantly reduced. [0003] The quantization is a common model compression and ...

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 Patents(China)
IPC IPC(8): G06N3/04G06T3/40
CPCG06T3/4053G06N3/045G06T3/40G06N3/04
Inventor 胡慧
Owner HUAWEI TECH CO LTD