Infrared image denoising method for gas leakage based on transfer learning

SA Yu; ZHANG Shilei; TAN Mei; ZHANG Yinghu; YANG Yunpeng; MA Xiangyun; LI Qifeng

doi:10.3969/j.issn.1673-6141.2024.05.004

Journal of Atmospheric and Environmental Optics, Volume. 19, Issue 5, 543(2024)

Infrared image denoising method for gas leakage based on transfer learning

SA Yu^1,2, ZHANG Shilei^1,2, TAN Mei^1,2, ZHANG Yinghu^1,2, YANG Yunpeng^1,2, MA Xiangyun^1,2、*, and LI Qifeng^1,2、**

Author Affiliations

¹School of Precision Instrument and Opto-electronics Engineering, Tianjin University, Tianjin 300072, China

²Tianjin Key Laboratory of Biomedical Detecting Techniques and Instruments, Tianjin 300072, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(11)

Fig. 1. Flowchart of the gas leak infrared image denoising method based on deep transfer learning

Download full size

View in Article

Fig. 2. CNN model of infrared image denoising method for gas leakage based on deep transfer learning

Download full size

View in Article

Fig. 3. Infrared images using four denoising algorithms. (a) Ideal image for the test set; (b) input image for the test set;(c) denoised image using CNN model; (d) denoised image using BF algorithm; (e) denoised image using NMF algorithm;(f) denoised image using WT algorithm

Download full size

View in Article

Fig. 4. Infrared images using four denoising algorithms. (a) Ideal image for the test set; (b) input image for the test set; (c) denoised image using CNN model; (d) denoised image using BF algorithm; (e) denoised image using NMF algorithm;(f) denoised image using WT algorithm

Download full size

View in Article

Fig. 5. The result of the gas infrared images based on four denoising algorithms. (a) PSNR of 10 gas infrared image samples;(b) SSIM of 10 gas infrared image samples

Download full size

View in Article

Fig. 6. Real gas infrared images and denoised images. (a) Four consecutive gas images taken with an uncooled infrared camera; time interval: 50 ms; (b) infrared images denoised using CNN model

Download full size

View in Article

Table 1. Parameters of operation modules in CNN model
View table
View in Article
Table 1. Parameters of operation modules in CNN model
Operation Module Layer Kernel Size Filters Activation
1-4 Conv2D $3 \times 3$ 64 ReLU
MaxPooling2D $2 \times 2$ — —
5-8 Conv2D $3 \times 3$ 64 ReLU
UpSampling $2 \times 2$ — —
9 Conv2D $3 \times 3$ 64 ReLU
10 Conv2D $3 \times 3$ 1 —

Table 2. Average PSNR and SSIM of infrared images based on four denoising algorithms
View table
View in Article
Table 2. Average PSNR and SSIM of infrared images based on four denoising algorithms
Method CNN BF NMF WT
PSNR/dB 38.45 21.32 23.17 24.56
SSIM 0.9521 0.3751 0.3978 0.4144

Table 3. Average PSNR and SSIM of gas infrared images based on four denoising algorithms
View table
View in Article
Table 3. Average PSNR and SSIM of gas infrared images based on four denoising algorithms
Method CNN BF NMF WT
PSNR/dB 37.89 16.35 20.61 17.20
SSIM 0.9485 0.1741 0.2432 0.1913

Table 4. Average PSNR of gas infrared images under different noise levels with four denoising algorithms
View table
View in Article
Table 4. Average PSNR of gas infrared images under different noise levels with four denoising algorithms
Method PSNR/dB
$σ = 20$ $σ = 40$ $σ = 60$ $σ = 80$ $σ = 100$
BF 16.21 15.98 15.13 14.77 13.54
NMF 20.32 19.81 19.27 18.44 17.01
WT 17.03 16.65 16.26 16.89 15.31
CNN 37.80 37.41 37.23 37.11 36.20

Table 5. Average SSIM of gas infrared images under different noise levels with four denoising algorithms
View table
View in Article
Table 5. Average SSIM of gas infrared images under different noise levels with four denoising algorithms
Method SSIM
$σ = 20$ $σ = 40$ $σ = 60$ $σ = 80$ $σ = 100$
BF 0.1739 0.1721 0.1687 0.1642 0.1594
NMF 0.2421 0.2412 0.2397 0.2356 0.2301
WT 0.1701 0.1682 0.1673 0.1621 0.1537
CNN 0.9475 0.9470 0.9462 0.9442 0.9426

Tools

Get Citation

Copy Citation Text

Yu SA, Shilei ZHANG, Mei TAN, Yinghu ZHANG, Yunpeng YANG, Xiangyun MA, Qifeng LI. Infrared image denoising method for gas leakage based on transfer learning[J]. Journal of Atmospheric and Environmental Optics, 2024, 19(5): 543

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category:

Received: Jul. 8, 2022

Accepted: --

Published Online: Jan. 8, 2025

The Author Email: Xiangyun MA (mxy1994@tju.edu.cn), Qifeng LI (qfli@tju.edu.cn)

DOI:10.3969/j.issn.1673-6141.2024.05.004

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Parameters of operation modules in CNN model

Table 1. Parameters of operation modules in CNN model

Table 2. Average PSNR and SSIM of infrared images based on four denoising algorithms

Table 2. Average PSNR and SSIM of infrared images based on four denoising algorithms

Table 3. Average PSNR and SSIM of gas infrared images based on four denoising algorithms

Table 3. Average PSNR and SSIM of gas infrared images based on four denoising algorithms

Table 4. Average PSNR of gas infrared images under different noise levels with four denoising algorithms

Table 4. Average PSNR of gas infrared images under different noise levels with four denoising algorithms

Table 5. Average SSIM of gas infrared images under different noise levels with four denoising algorithms

Table 5. Average SSIM of gas infrared images under different noise levels with four denoising algorithms