Eureka AIR delivers breakthrough ideas for toughest innovation challenges, trusted by R&D personnel around the world.

Small target detection method based on multilevel residual network perception and attention mechanism

A small target detection and network perception technology, applied in the field of small target detection based on multi-level residual network perception and attention mechanism, can solve the problems of few training samples, difficult positioning, missing detection of small target objects, etc., and achieve enhanced semantics Information and positioning information, improve generalization ability, good detection effect

Active Publication Date: 2022-07-29
SHANDONG ARTIFICIAL INTELLIGENCE INST +3
View PDF5 Cites 6 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] The present invention provides a small target detection method based on a multi-level residual network perception and attention mechanism, which solves the problem of missed detection of small target objects, false detection, low resolution, few available features, and few training samples in traditional target detection methods , little contribution to the loss function, difficult positioning and many other problems

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
  • Small target detection method based on multilevel residual network perception and attention mechanism
  • Small target detection method based on multilevel residual network perception and attention mechanism
  • Small target detection method based on multilevel residual network perception and attention mechanism

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0060] like figure 1 As shown, it is the operation flow chart of the small target detection method based on the multi-level residual network perception and attention mechanism of the present invention. The details of the implementation steps of the method are as follows:

[0061] 1) Selective sample replication to enhance and expand the training set, the specific operations are as follows:

[0062] Since the number of samples in the training data set is small and the target is small, the generalization ability of the model will be poor after training, and the model cannot fit the target data well. Selective sample replication enhancement can better solve the above problems; the design of the present invention is selective The sample copy enhancement is different from the previous Copy-Paste. The design of the present invention randomly enlarges the target sample area less than 500 pixels to 1.5-2 times of the original image, the sample area between 500 and 1,000 pixels is ran...

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 small target detection method based on multilevel residual network perception and an attention mechanism. A single-stage detector YOLOv5 is adopted in Baseline; the method comprises the following specific steps: (1) constructing a virtual training sample, expanding the scale of a training data set, and improving the overall performance of a model; (2) extracting multi-dimensional features of the image by using multi-layer residual convolution; (3) an attention mechanism is used for enhancing a superficial layer feature map; (4) connecting the feature pyramid with the shallow feature map; (5) using a loss function optimization model to predict target position information, category information and confidence; and (6) the P2 detection layer is matched with Lufl and VIoU Loss to predict a small target. According to the method, the data set is enhanced and expanded through selective samples, the difference between the samples is balanced, the shallow feature map is fully utilized, and the recall rate and the accuracy rate of small target detection are remarkably improved.

Description

technical field [0001] The invention relates to the field of target detection and recognition, and relates to a small target detection method based on a multi-level residual network perception and attention mechanism. Background technique [0002] At present, different scenarios have different definitions for small goals, and a unified standard has not yet been formed. The existing mainstream small target definition methods are mainly divided into the following two categories, namely, the definition based on relative scale and the definition based on absolute scale. Relative scale is generally defined as the ratio of the bounding box area to the image area with a median between 0.08% and 0.58%. Absolute scale is generally defined as an object with a resolution less than 32 pixels by 32 pixels. The design of existing algorithms tends to focus more on the detection performance of large and medium-scale objects. There are not many optimization designs for the characteristics...

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): G06V10/774G06V10/82G06N3/04G06N3/08
CPCG06V10/774G06V10/82G06N3/08G06V2201/07G06N3/045
Inventor 高赞纪威王水跟徐国智顾竟潇刘大扬郝敬全
Owner SHANDONG ARTIFICIAL INTELLIGENCE INST
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
Eureka Blog
Learn More
PatSnap group products