Target detection active sampling method based on time sequence variance threshold

A technology of target detection and active sampling, which is applied in the direction of instruments, character and pattern recognition, computer components, etc., can solve the problems of missing key information and high cost, and achieve the effect of reducing negative impact and labeling cost

Active Publication Date: 2022-04-12
NANJING UNIV OF AERONAUTICS & ASTRONAUTICS
View PDF3 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The labeling cost of the former is too high, while the sampling of the latter is random, and key information may be lost

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
  • Target detection active sampling method based on time sequence variance threshold
  • Target detection active sampling method based on time sequence variance threshold
  • Target detection active sampling method based on time sequence variance threshold

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0041] Such as figure 1 Shown is a flow diagram of the mechanism of the present invention. Assume that initially there is a dataset consisting of a small number of annotated images , and a dataset consisting of a large number of unlabeled images . First, initialize the target detection model with a small amount of labeled time-series image data set L and set the number of planned query samples n and the variance threshold δ, and initialize the number of selected samples q to 0. Subsequently, the model is initialized with the labeled image set, and the target detection model is output to predict the unlabeled frame. Then, calculate the model uncertainty of each iteration for each unlabeled frame according to the prediction results, and arrange them from large to small. Take the sample with the largest model uncertainty, if the timing variance of the sample is greater than the threshold and the adjacent frame is not selected, then the sample will be marked by the expert qu...

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 target detection active sampling method based on a time sequence variance threshold. Comprising the following steps: 1, collecting a large amount of unlabeled time sequence data and a small amount of labeled data; 2, setting a query sample number n and a variance threshold value delta; 3, initializing the model; 4, outputting a prediction result of the unlabeled frame by the target detection model; 5, calculating the model uncertainty of each iteration for the unlabeled frame according to the prediction result; 6, a sample with the maximum model uncertainty is taken, and if the time sequence variance is larger than a threshold value and adjacent frames are not selected, the sample is marked to query; 7, updating the labeled image set, the unlabeled image set and the prediction model; and 8, returning to the step 4 or querying enough samples and outputting a target detection model f. For a target detection task of time series data in an automatic driving scene, a special active learning index is set to reduce the labeling cost.

Description

technical field [0001] The invention belongs to the technical field of digital image automatic labeling, and in particular relates to an active sampling method for target detection based on time series variance threshold. Background technique [0002] In the actual industrial application process, data has always been valued as the core resource of the Industrial Internet. Massive data is generated after the interconnection of people, machines, and things, which promotes the development of the industry. However, these data usually contain a large amount of redundant data, and data extraction and cleaning have become a top priority. One of the prominent areas is the timing data of autonomous driving. The autonomous driving data is usually intercepted from multiple driving segments recorded by the camera. Each segment contains continuous pictures in the driving scene, that is, sequential frames. For this type of data, using active learning algorithm for selective sampling can ...

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): G06V20/56G06V10/778
Inventor 黄圣君罗世发
Owner NANJING UNIV OF AERONAUTICS & ASTRONAUTICS
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