Convolutional neural network structure optimization method and device and electronic equipment

A convolutional neural network and network structure technology, applied in neural learning methods, biological neural network models, neural architectures, etc., can solve the problems of consuming large computing and storage resources, difficult to deploy mobile devices, and large number of model parameters. Achieve the effect of reducing memory and computing power consumption, saving hardware computing resources, and small size

Pending Publication Date: 2020-04-21
盛景智能科技嘉兴有限公司
View PDF0 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, in some applications, such as face detection, it is usually necessary to design a large-scale convolutional neural network to ensure the accuracy of the model, resulting in a very large number of parameters in the model, which consume a large amount of computing and storage resources, and consume computing power often exceeds expectations
Even some models tend to be over a few hundred megabytes, which makes it difficult to deploy on mobile devices

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 structure optimization method and device and electronic equipment
  • Convolutional neural network structure optimization method and device and electronic equipment
  • Convolutional neural network structure optimization method and device and electronic equipment

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0035] see figure 1 , the embodiment of the present invention provides a convolutional neural network structure optimization method, the method comprising:

[0036] Step S102, within the first preset accuracy range, reduce the number of layers of the pre-trained convolutional neural network and the number of channels of each layer to obtain the first convolutional neural network;

[0037] Step S104, optimizing the first convolutional neural network according to the cascaded multi-channel network structure to obtain a second convolutional neural network;

[0038] Step S106, within the second preset accuracy range, reduce the number of layers of the second convolutional neural network and the number of channels of each layer to obtain a third convolutional neural network, so as to apply the third convolutional neural network The internet.

[0039] For step S102, within the first preset accuracy range, the number of layers of the pre-trained convolutional neural network and the...

Embodiment 2

[0061] refer to image 3 , on the basis of the above-mentioned embodiments, the embodiment of the present invention also provides another convolutional neural network structure optimization method, which is different from the above-mentioned embodiment 1 in that the method also includes:

[0062] Step S202, optimizing the third convolutional neural network by using a preset quantization technique of the deep neural network.

[0063] Quantization here is an optimization method for converting floating-point calculations into fixed-point calculations, including 8-bit quantization and less-bit quantization (4 bits, 2 bits, etc.).

[0064] The above-mentioned preset quantization technology of the deep neural network includes quantization time and quantization size. The quantization time includes quantization after complete model training (after completion) and quantization during model training. The quantization size refers to the preset number of bits (for example, the above-menti...

Embodiment 3

[0069] Based on the same inventive concept, the embodiment of the present application also provides a convolutional neural network structure optimization device corresponding to the convolutional neural network structure optimization method. Since the problem-solving principle of the device in the embodiment of the present application is the same as that of the above volume in the embodiment of the present application The method for optimizing the structure of the product neural network is similar, so the implementation of the device can refer to the implementation of the method, and the repetition will not be repeated.

[0070] Figure 4 It is a schematic diagram of a convolutional neural network structure optimization device provided in an embodiment of the present application.

[0071] refer to Figure 4 , the device includes: a first optimization module 401, a second optimization module 402 and a third optimization module 403;

[0072] Wherein, the first optimization mod...

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 embodiment of the invention provides a convolutional neural network structure optimization method and device and electronic equipment, and relates to the field of artificial intelligence. The method comprises the steps: reducing the number of layers of a pre-trained convolutional neural network and the number of channels of each layer in a first preset precision range, and obtaining a first convolutional neural network; optimizing the first convolutional neural network according to a cascaded multi-path network structure to obtain a second convolutional neural network; and in a second preset precision range, reducing the number of layers of the second convolutional neural network and the number of channels of each layer to obtain a third convolutional neural network so as to apply thethird convolutional neural network. According to the method, on the premise of ensuring the precision, the parameter quantity of the network can be greatly reduced, the consumption of a memory and calculation power is reduced in the calculation process, and more hardware calculation resources are saved, so that the effect of overall acceleration is achieved; according to the method, the size of the model is smaller, and the operation speed is higher.

Description

technical field [0001] The present invention relates to the field of target detection, in particular to a convolutional neural network structure optimization method, device and electronic equipment. Background technique [0002] At present, convolutional neural networks have been widely used in various fields of artificial intelligence (AI), including human detection, face recognition, segmentation, pose estimation, etc., and with the application of AI in the industry Extensive and deep into all aspects of life and production. However, in some applications, such as face detection, it is usually necessary to design a large-scale convolutional neural network to ensure the accuracy of the model, resulting in a very large number of parameters in the model, which consume a large amount of computing and storage resources, and consume The computing power often exceeds expectations. Even some models are often over a few hundred megabytes, which makes it difficult to deploy on mobi...

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): G06N3/04G06N3/08
CPCG06N3/082G06N3/045
Inventor 唐振赵小伟王豪
Owner 盛景智能科技嘉兴有限公司
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