Unlock instant, AI-driven research and patent intelligence for your innovation.

A low-light target detection method based on ms-wsda

A target detection and low-illumination technology, applied in target detection technology, weak supervision and domain adaptation, and image enhancement, can solve problems such as large amount of calculation, sensitivity to the size and number of anchor frames, and imbalance between positive and negative samples

Active Publication Date: 2022-05-13
GUILIN UNIV OF ELECTRONIC TECH
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Most detectors with anchors have the disadvantages of being sensitive to the size and number of anchor boxes, unbalanced positive and negative samples, and large amount of calculation.

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
  • A low-light target detection method based on ms-wsda
  • A low-light target detection method based on ms-wsda
  • A low-light target detection method based on ms-wsda

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0055]A low-light target detection method based on MS-WSDA, comprising the steps of:

[0056] 1) Integrating data sets: including:

[0057] 1-1) Select the images in the PASCAL VOC2007 data set. The PASCAL VOC2007 data set has 5011 images in the training set and 4952 images in the test set, a total of 9963 images, including 20 types. The PASCAL VOC2007 data set is used for PL-AFD Pre-training, Table 1 is the source of the data set:

[0058] Table 1

[0059]

[0060]

[0061] 1-2) Select the SID dataset. The SID dataset contains 5094 low-illumination images and their corresponding normal-illuminance images. Randomly select 70% of the images as the training set images and 30% of the images as the test-set images. Among them, the normal illumination The image is used to test the pre-trained pixel-based anchor-free detector PL-AFD to generate pseudo-labels. The training set of low-light images and normal-light images is used to train the low-light enhancement network LLENe...

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 low-illuminance target detection method based on multi-scale weak supervision and domain adaptation, including the following steps: 1) integrating data sets; Label generation; 3) training of the low-light image enhancement network LLENet; 4) training of the domain adaptation module; 5) training of the self-supervised module; 6) testing of the entire low-light object detection network. This method can bridge the pixel-level and semantic-level gap between low-illumination images and normal-illumination images, and improve the detection accuracy of object detectors for low-illumination images.

Description

technical field [0001] The present invention relates to the fields of image enhancement, target detection technology, weak supervision and domain adaptation, in particular to a multi-scale weak supervision and domain adaptation MS-WSDA (Multi-Scale Weakly Supervised and Domain Adaptive, referred to as MS-WSDA) low-illumination target Detection method. Background technique [0002] Most of the existing object detectors are used to detect images under normal illumination, but the detection effect on low-illumination images is extremely poor. This is because low-illumination images have low contrast, blurred content, loss of details and other distracting factors, making it difficult for object detectors to extract salient features. In recent years, the application of low-illuminance enhancement technology has effectively improved image illumination and enhanced human subjective visual experience of images. According to the characteristics of existing methods, low-light enhanc...

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 Patents(China)
IPC IPC(8): G06V10/80G06V10/774G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06N3/045G06F18/214G06F18/253
Inventor 江泽涛李慧
Owner GUILIN UNIV OF ELECTRONIC TECH