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

Convolution operation based matrix conversion method and apparatus, and medium

A technology of convolution operation and transformation device, applied in the field of convolutional neural network, can solve the problems of multiple memory resources, low memory use efficiency, large scale, etc., to reduce the occupation of memory resources, reduce the overall number of repetitions, and improve the use efficiency Effect

Inactive Publication Date: 2018-06-15
ZHENGZHOU YUNHAI INFORMATION TECH CO LTD
View PDF0 Cites 10 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, in actual usage, the number of rows and columns of the feature matrix is ​​often much larger than the number of rows and columns of the convolution kernel matrix. Therefore, the resulting feature matrix obtained by expanding the feature matrix in IM2COL mode is relatively large and will occupy more memory resources. , which leads to low memory usage efficiency and affects the efficiency of the overall convolution operation

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
  • Convolution operation based matrix conversion method and apparatus, and medium
  • Convolution operation based matrix conversion method and apparatus, and medium
  • Convolution operation based matrix conversion method and apparatus, and medium

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0044] figure 2 It is a flow chart of a convolution operation-based matrix conversion method provided by an embodiment of the present invention. Please refer to figure 1 , the specific steps of the matrix transformation method based on convolution operation include:

[0045] Step S10: Obtain the original feature matrix and convolution kernel matrix.

[0046] It should be noted that both the original feature matrix and the convolution kernel matrix are parameters required for convolution operations. Since the convolution operation is often used in the convolutional neural network to analyze the image, the original feature matrix can be converted from the image, and the clarity of the image directly determines the number of rows and columns of the original feature matrix.

[0047] Step S11: Count the total number of rows of the original feature matrix, the number of kernel rows of the convolution kernel matrix, and the number of kernel columns of the convolution kernel matri...

Embodiment 2

[0054] On the basis of the foregoing embodiments, as a preferred implementation manner, the target number of rows is equal to the total number of rows.

[0055] The following diagrams are used to illustrate, please refer to Figure 4 .

[0056] Figure 4 It is a schematic diagram of another expansion of the original feature matrix in this scheme. Such as Figure 4 As shown in , the convolution operation is performed by using the 5*5 original feature matrix and the 3*3 convolution kernel matrix as an example. A in the figure is the original feature matrix, B is the convolution kernel matrix, and C is the result feature matrix. Among them, the result feature matrix is ​​formed by expanding and combining the target matrices. The specific expansion method of the target matrix is ​​the same as the existing expansion method, and it can also be reflected according to the labels of the elements in the matrix in the figure, so I will not repeat them here. .

[0057] image 3 and...

Embodiment 3

[0077] The embodiment of a matrix conversion method based on convolution operation has been described in detail above, and the present invention also provides a matrix conversion device based on convolution operation, because the embodiment of the device part and the embodiment of the method part Corresponding to each other, so for the embodiment of the device part, please refer to the description of the embodiment of the method part, and details will not be repeated here.

[0078] Figure 6 A structural diagram of a matrix conversion device based on convolution operation provided by an embodiment of the present invention. Such as Figure 6 As shown, the matrix conversion device based on convolution operation provided by the embodiment of the present invention includes:

[0079] The matrix acquisition module 10 is used to acquire the original feature matrix and the convolution kernel matrix.

[0080] The statistical module 11 is used to count the total number of rows of the...

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 invention discloses a convolution operation based matrix conversion method and apparatus, and a medium. The method comprises the steps: acquiring an original feature matrix and a convolution kernel matrix; collecting statistics of the total number of rows of the original feature matrix, the number of kernel rows of the convolution kernel matrix, and the number of kernel columns of theconvolution kernel matrix; sequentially selecting the target matrix with the target number of rows and the target number of kernel columns in the original feature matrix, and expanding the original feature matrix to be combined into a result feature matrix, wherein the number of target rows is greater than the number of kernel rows and is less than or equal to the total number of rows; and writingthe result feature matrix to a memory to carry out the convolution operation with the convolution kernel matrix. According to the method disclosed by the present invention, the size of the result feature matrix obtained after the original feature matrix is expanded is smaller; and the occupation of memory resources during the convolution operation is relatively reduced, so that the memory usage efficiency is reduced, and the efficiency of the overall convolution operation is ensured. The present invention further provides a convolution operation based matrix conversion apparatus, and a medium, which have the above described beneficial effects.

Description

technical field [0001] The present invention relates to the field of convolutional neural networks, in particular to a matrix conversion method, device and medium based on convolution operations. Background technique [0002] Convolutional neural network is an efficient recognition method that has been developed in recent years and has attracted widespread attention. [0003] The difference between the convolutional neural network and the ordinary neural network is that the convolutional neural network contains a feature extractor with a convolutional layer and a sub-sampling layer structure. In the convolutional layer of the convolutional neural network, there are usually several feature planes, and each feature plane is composed of neurons arranged in a matrix. The neurons of the same feature plane share the same convolution kernel, and the convolution kernel is usually random. The form of the decimal matrix is ​​initialized, and the convolutional neural network often inv...

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): G06N3/063G06F17/16G06F17/15
CPCG06N3/063G06F17/153G06F17/16
Inventor 于福海吴韶华
Owner ZHENGZHOU YUNHAI INFORMATION TECH CO LTD