Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Few-sample target detection method based on singular value decomposition feature enhancement

A singular value decomposition and target detection technology, applied in the field of few-sample target detection based on singular value decomposition feature enhancement, can solve the problems of difficult generalization of features, difficult target classification and positioning, and weak discriminative ability, so as to improve representativeness, Improve generalization and discrimination, and improve the effect of positioning and classification accuracy

Pending Publication Date: 2022-01-25
XIDIAN UNIV
View PDF0 Cites 3 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, in the existing few-sample target detection methods, the features extracted in the feature extraction are difficult to generalize, and the discriminative ability is weak, and the learned model does not have good generalization and discriminative ability, and it is difficult to distinguish the features in the image. Accurate classification and positioning of targets

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
  • Few-sample target detection method based on singular value decomposition feature enhancement
  • Few-sample target detection method based on singular value decomposition feature enhancement
  • Few-sample target detection method based on singular value decomposition feature enhancement

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0039] In the existing few-sample target detection methods, due to the small number of samples, the extracted features cannot well represent the features of this category, and the learned model has poor generalization and discrimination. The present invention conducts research on the above problems, and provides a few-sample target detection method based on singular value decomposition feature enhancement.

[0040] The present invention is a few-sample target detection method based on singular value decomposition feature enhancement, see figure 1 , figure 1 It is a flow chart of the present invention, and the few-sample target detection method based on singular value decomposition feature enhancement of the present invention comprises the following steps:

[0041] (1) Obtain the image data set for target detection: the image data set includes a training sample set and a test sample set. The commonly used data sets for few-sample target detection are PASCAL VOC and COCO data s...

Embodiment 2

[0068] The few-sample target detection method based on singular value decomposition feature enhancement is the same as embodiment 1, and the construction feature enhancement module described in step (3) carries out feature enhancement and includes the following steps:

[0069] 3.1) Singular value decomposition is performed on the feature map: the feature map F extracted in step (2), rescaled to Singular value decomposition is performed on the adjusted feature map F to obtain where U is the left singular matrix after singular value decomposition, V is the right singular matrix after singular value decomposition, U and V are orthogonal unitary matrices, Σ is an m×n diagonal matrix sorted by size of the singular values ​​of the diagonal, its dimension is 2k, m is the number of channels of the feature map F, and ω is the width of the feature map F , h is the height of the feature map F, n is the dimension of the adjusted feature map F, n=ω×h.

[0070] 3.2) Learning genera...

Embodiment 3

[0080] The few-sample target detection method based on singular value decomposition feature enhancement is the same as embodiment 1-2, two kinds of feature maps described in step (6) are fused to form a feature fusion layer, including the following steps:

[0081] 6.1) Generalized feature map fusion of advanced discriminant information: The generated feature map P of the candidate frame area is coded and represented by the codeword set learned by dictionary learning, expressed as Rep, and then the representation Rep is combined with the generated feature map P of the candidate frame area Alignment is performed through the RoI alignment layer. After alignment, the two are merged to achieve feature fusion through matrix splicing, and a generalized feature map [φ(P),φ(Rep)] with fusion features is obtained. The representation process is as follows:

[0082]

[0083] Among them, ψ is a fully connected layer that maps the feature map P to the dictionary space, p is the feature su...

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 provides a few-sample target detection method based on singular value decomposition feature enhancement. The problem that a few-sample target detection method is poor in generalization and discrimination is solved. The method comprises the steps of obtaining a target detection image data set; performing feature extraction on the training sample set image; constructing a feature enhancement module to enhance the extracted features; enabling the RPN module to generate a candidate frame area and carrying out RoI alignment; fusing the two feature maps to form a feature fusion layer; positioning and classifying the frame of the target object; carrying out training on the improved Faster R-CNN network; and performing target detection on the to-be-detected image. According to the method, three parts of a feature enhancement module, a feature fusion layer and an Lkl loss function are provided, more essential features of an image and discrimination information in a high-dimensional space are learned, the features have good generalization and discrimination, the positioning and classification precision of few-sample target detection is effectively improved, and the method can be used in the fields of robot navigation, intelligent video monitoring and the like.

Description

technical field [0001] The invention belongs to the technical field of computer vision image detection, and in particular relates to image few-sample target detection, and specifically a few-sample target detection method based on singular value decomposition feature enhancement, which can be used in the fields of robot navigation, intelligent video monitoring and the like. Background technique [0002] Object detection is an image segmentation based on the geometric and statistical characteristics of the object, which can locate and recognize the object at the same time. Existing object detection methods are mainly divided into two categories: two-stage methods based on RPN and single-stage methods based on SSD and YOLOv1 / v2 / v3. The RPN-based two-stage method focuses on improving the accuracy of target detection, such as RCNN, Fast-RCNN, Faster-RCNN, Mask RCNN, etc. The single-stage method based on SSD and YOLOv1 / v2 / v3 focuses on improving the speed of target detection, su...

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): G06V40/10G06V10/44G06V10/764G06V10/82G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06N3/047G06N3/045G06F18/2415G06F18/241
Inventor 杨延华孙娜武阿明杨木李王宇宣邓成
Owner XIDIAN UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com