Deep learning-based dual-supervision image defogging method, system, medium and equipment

A deep learning, dual-supervised technology, applied in the field of image processing, can solve problems such as affecting the dehazing effect, and achieve the effect of avoiding gradient explosion, avoiding subjective settings, and making up for insufficient prior knowledge.

Active Publication Date: 2019-08-06
JINAN UNIVERSITY
View PDF2 Cites 29 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

When performing image processing based on image restoration, if you do not use external equipment to measure the image depth of field, you need to use methods such as dark channel prior or color decay prior to estimate the

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
  • Deep learning-based dual-supervision image defogging method, system, medium and equipment
  • Deep learning-based dual-supervision image defogging method, system, medium and equipment
  • Deep learning-based dual-supervision image defogging method, system, medium and equipment

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0070] In this embodiment, a double-supervised lightweight image defogging system based on deep learning is provided, including: a foggy image and a fogless image sample acquisition module, a neural network system building module and a neural network system training module;

[0071] The foggy image and fog-free image sample acquisition module is used to acquire foggy image and fog-free image samples, and the fog-free image samples are used as label comparison samples;

[0072] Such as figure 1 As shown, the neural network system building block is used to construct the neural network system, and the neural network system includes a down-sampling module, a feature extraction module and an up-sampling module;

[0073] Downsampling module: The downsampling module includes a convolutional layer and a maximum pooling layer. The convolutional layer extracts image features, and then uses the maximum pooling operation. Its function is to reduce the size of the image, so that subsequent...

Embodiment 2

[0084] Such as figure 2 As shown, the present embodiment provides a double-supervised image defogging method based on deep learning, comprising the following steps:

[0085] S1: Obtain foggy images and tag non-foggy images;

[0086] S2: Build a neural network system, initialize the convolution kernel weights of the neural network to a Gaussian random distribution, and set the deviation value; in this embodiment, initialize the convolution kernel weight w in the neural network to a mean value of 0 and a variance of Gaussian random distribution of 0.05, the deviation b is set to a constant of 0.1;

[0087] The specific construction steps of the neural network system are as follows:

[0088] S21: Downsampling: Extract image features through the convolutional layer, extract the most significant features of the image through the maximum pooling layer, and reduce the image size, so that subsequent network calculations can be completed faster;

[0089] S22: Feature extraction: ad...

Embodiment 3

[0130] This embodiment also provides a storage medium, which can be a storage medium such as ROM, RAM, magnetic disk, optical disk, etc., and the storage medium stores one or more programs. When the programs are executed by the processor, the implementation of embodiment 2 based on A double-supervised image defogging method for deep learning, the method includes the following steps:

[0131] S1: Obtain foggy images and tag non-foggy images;

[0132] S2: Build a neural network system, initialize the convolution kernel weights of the neural network to a Gaussian random distribution, and set the deviation value; in this embodiment, initialize the convolution kernel weight w in the neural network to a mean value of 0 and a variance of Gaussian random distribution of 0.05, the deviation b is set to a constant of 0.1;

[0133] The specific construction steps of the neural network system are as follows:

[0134] S21: Downsampling: Extract image features through the convolutional la...

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 deep learning-based dual-supervised lightweight image defogging method, system, medium and equipment. The method comprises the following steps of: obtaining a foggy image anda label fog-free image; constructing a neural network system; training the neural network system; inputting the foggy image into the neural network system to obtain a first transmission image, and performing an image restoration algorithm on the foggy image to obtain a second transmission image; performing mean square error calculation on the first transmission map and the second transmission mapto obtain a loss function Lt; performing atmospheric scattering model inverse operation on the first transmission image to obtain a defogged image; comparing the defogged image with the label fog-free image to obtain a loss function Ld; combining Lt and Ld according to a set proportion to obtain Ltotal; inputting the foggy image into the trained neural network system to obtain a defogged image. The system comprises a foggy and non-foggy image sample acquisition module, a neural network system construction module and a neural network system training module. The network parameter magnitude is small, the training time is short, subjective setting can be avoided, and the defogging effect is enhanced.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a double-supervised lightweight image defogging method, system, medium and equipment based on deep learning. Background technique [0002] With the intensification of environmental pollution, haze weather is becoming more and more frequent. Due to the scattering effect of suspended particles in the air (such as fog, haze, etc.), bad weather not only leads to low visibility, but also images taken in hazy weather often have degradation problems such as low contrast, color shift, and poor visual effects. . Most outdoor vision systems need to extract image features clearly and accurately, and the degradation of image quality will affect the effectiveness of subsequent computer vision tasks. [0003] In the early research, the method based on image enhancement was adopted, which only improved the contrast of the haze image without considering the physical model. Th...

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): G06T5/00
CPCG06T5/003G06T2207/20081G06T2207/20084
Inventor 李展陈昱铃黄维健钟锐彬张建航
Owner JINAN UNIVERSITY
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