Multi-scale feature extraction and fusion method and device

A technology of multi-scale features and fusion methods, applied in neural learning methods, instruments, biological neural network models, etc.

Pending Publication Date: 2020-09-15
QILU UNIV OF TECH
View PDF0 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] The technical task of the present invention is to address the above deficiencies and provide a multi-scale feature extraction and fusion method and device to solve the problem of how to achieve multi-scale feature extraction and fusion under the premise of reducing noise and redundant data

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
  • Multi-scale feature extraction and fusion method and device
  • Multi-scale feature extraction and fusion method and device
  • Multi-scale feature extraction and fusion method and device

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0138] like figure 1 and figure 2 As shown, a multi-scale feature extraction and fusion method of the present invention uses multiple different Faster R-CNN deep learning networks to perform multi-scale feature extraction to obtain multiple groups of different feature sets, and the above-mentioned multiple groups of different feature sets For DCA feature fusion, each feature set includes multiple feature vectors with the same dimension.

[0139] The method comprises the steps of:

[0140] S100. Input the original picture into each learning network, each learning network is a Faster R-CNN deep learning network, including a convolutional layer network, an RPN network, and a Fast R-CNN network;

[0141] S200. In each learning network, the original image is extracted through the convolutional layer to obtain a feature map, and the target detection and precise positioning are performed on the feature map through the RPN network to obtain candidate frames, and RoI pooling in the ...

Embodiment 2

[0235] A multi-scale feature extraction and fusion device of the present invention includes a feature extraction module and a feature fusion module. The feature extraction module is used to perform multi-scale feature extraction through a plurality of different Faster R-CNN deep learning networks. Each learning network is It is a Faster R-CNN deep learning network, including a convolutional layer network, an RPN network, and a Fast R-CNN network. In each learning network, the original image is extracted through the convolutional layer to obtain a feature map, and the feature map is obtained through the RPN network. Perform target detection and precise positioning to obtain candidate frames, and perform the maximum pooling operation on the candidate frames through the RoI pooling layer in the Fast R-CNN network, and output a set of feature sets including multiple feature vectors with the same dimension; feature fusion module It is used to perform DCA feature fusion on multiple s...

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 multi-scale feature extraction and fusion method and device, belongs to feature extraction and fusion, and aims to solve the technical problem of how to realize multi-scale feature extraction and fusion on the premise of reducing noise and redundant data. According to the method, multi-scale feature extraction is carried out through a plurality of Faster R-CNN deep learning networks to obtain a plurality of groups of different feature sets, DCA feature fusion is carried out on the plurality of groups of different feature sets, and each feature set comprises a plurality of feature vectors with the same dimension. The device is used for carrying out multi-scale feature extraction and fusion through the multi-scale feature extraction and fusion method. The device comprises a feature extraction module and a feature fusion module. Compared with a conventional feature extraction and feature fusion method, the method is better in robustness, improves the target detection accuracy, and solves a problem that a single-scale feature may cause a detection error.

Description

technical field [0001] The invention relates to the field of feature extraction and fusion, in particular to a multi-scale feature extraction and fusion method and device. Background technique [0002] Object detection is an important research content of computer vision, which is both a challenge and a difficult problem. Although the traditional R-CNN target detection algorithm effectively improves the recognition accuracy compared with the previous target detection algorithm, the algorithm needs to select thousands of regions of interest for a picture, and each region of interest must be put into the convolution Network recognition, the amount of calculation is undoubtedly so huge that it cannot be practically used; Fast R-CNN overcomes the computational redundancy problem caused by R-CNN extracting deep features, and unifies feature extraction, target classification and frame regression into one structure However, the extraction process of the target candidate domain is s...

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): G06K9/46G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06V10/462G06N3/045G06F18/2415G06F18/25
Inventor 董爱美郑秋玉李志刚
Owner QILU UNIV OF TECH
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