A Denoising Method for Complex Background Noise of Infrared Imaging Guidance System Based on Deep Learning and Dual-domain Fusion

doi:10.12382/bgxb.2023.0307

Abstract

Abstract:

The infrared imaging guidance system is severely affected by harsh operating environments, and the imaging process is accompanied by complex background noise interference, which seriously affects the guidance and tracking accuracy of the system. In order to reduce the impact of composite noise on the infrared imaging effect, a priori setting of noise characteristics based on additive and multiplicative components is proposed based on the analysis of the causes and characteristics of various common noises; further, a denoising method for different types of noise based on deep convolutional neural network is designed according to the dual-domain fusion denoising idea of spatial domain and transform domain. This method introduces a rich gradient flow convolutional module into the UNet++ structure to reduce the gradient information redundancy and enhance the multi-receptive field feature extraction ability. A dimension attention mechanism is proposed to achieve dual-domain noise estimation according to the noise morphology characteristics. The high-order dual-tree complex wavelet transform is introduced as the domain transformation method to improve the recognition ability of noise components at different scales and directions. The effectiveness and superiority of the noise prior setting and dual-domain fusion denoising idea are verified through ablation experiments, and the proposed method demonstrates the excellent denoising ability for various types of noise through comparative experiments. The proposed method achieves the peak signal-to-noise ratio of 29.57 and the structural similarity index of 0.85 for Gaussian noise removal, which is superior to other typical noise suppression methods. For multi-type mixed noise, it achieves the denoising levels of 27.84 and 0.82, respectively. Moreover, the proposed method was validated to possess significant capability in removing the noises from real image.

Key words: infrared imaging, image denoising, deep learning, spatial domain denoising, transform domain denoising

CLC Number:

TP391

LI Ping, ZHOU Yu, CAO Ronggang, LI Fadong, CAO Yuxi, LI Jiawu, ZHANG Anqi. A Denoising Method for Complex Background Noise of Infrared Imaging Guidance System Based on Deep Learning and Dual-domain Fusion[J]. Acta Armamentarii, 2024, 45(6): 1747-1760.

Figures/Tables 17

Fig.1 Flow chat of the proposed denoising algorithm

Fig.2 Spatial domain noise estimation module

Fig.3 RGC module and its composition

Fig.4 Dimension-wise attention mechanism

Fig.5 Transform domain noise estimation module

Fig.6 Principle of 2D-DTCWT for image

Fig.7 Principle of dual-domain fusion module

Fig.8 Evolution process of training loss function values for three denoising models

Table 1 Average denoising performances of three denoising models for Gaussian noise of validation set samples

去噪模型	评估指标
去噪模型	PSNR	SSIM
变换域去噪模型	26.13	0.77
空间域去噪模型	27.24	0.81
双域融合去噪模型	29.57	0.85

Fig.9 Comparison of dual-domain fusion denoising effects on Gaussian noise

Fig.10 Visualization of denoising process

Table 2 Denoising performance of the proposed algorithm for different types of noise

噪声类型	噪声参数	PSNR	SSIM
条纹噪声	像素值随机变化范围[20, 50]	34.32	0.92
网格噪声	像素值随机变化范围[20, 50]	33.31	0.91
高斯噪声	服从高斯分布(σ=50)	29.57	0.85
泊松噪声	服从泊松分布(λ=60)	29.12	0.84
混合噪声	网格噪声+高斯噪声+泊松噪声	27.84	0.82

Fig.11 Denoising effects of the proposed algorithm for various types of noise

Fig.12 Denoising effects of varied denoising algorithms for Gaussian noise

Tab.3 Comparion of denoising performances (PSNR) of the proposed algorithm and typical algorithms across multiple datasets for Gaussian noise

去噪算法	PSNR
去噪算法	Set12 数据集	DND 数据集	BSD68 数据集	本文数据集
VisuShrink	22.74	21.59	21.83	22.18
BayesShrink	23.47	22.77	22.42	24.21
BM3D	25.82	24.45	24.81	26.09
DnCNN	28.68	25.95	25.23	26.59
FFDNet	28.65	26.01	26.35	27.37
DeamNet	28.12	26.56	26.77	28.21
SCUNet	28.72	28.05	27.51	29.33
DRUNet	27.90	26.41	26.59	28.34
本文算法	28.85	28.12	27.44	29.57

Tab.4 Comparison of computational complexities of the proposed denoising algorithm and other typical deep neural network denoising algorithms

去噪算法	参数量(Params)	计算量(FLOPs)
DnCNN	1.2×10⁶	309.8×10⁹
FFDNet	1.1×10⁶	68.5×10⁹
DRUNet	32.6×10⁶	554.5×10⁹
SCUNet	17.9×10⁶	319.1×10⁹
DeamNet	1.8×10⁶	582.8×10⁹
本文算法	0.8×10⁶	11.2×10⁹

Fig.13 Denoising effect of the proposed algorithm for real noise

References 26

[1]	张宇飞, 李文豪, 王东振. 某型红外光学导引系统改进设计研究[J]. 航空维修与工程, 2021, 366(12): 68-71.
	ZHANG Y F, LI W H, WANG D Z. Improved design of an infrared optical guidance system[J]. Aviation Maintenance & Engineering, 2021, 366(12): 68-71. (in Chinese)
[2]	魏政, 杜勇, 刘辉, 等. 多模复合制导技术的发展现状与分析[J]. 航空兵器, 2022, 29(6): 26-33.
	WEI Z, DU Y, LIU H, et al. Development status and analysis of multi-mode composite guidance technology[J]. Aero Weaponry, 2022, 29(6): 26-33. (in Chinese)
[3]	卢晓, 梁晓庚, 贾晓洪. 红外成像空空导弹智能化抗干扰研究[J]. 红外与激光工程, 2021, 50(4): 20200240.
	LU X, LIANG X G, JIA X H. Study on intelligent counter-countermeasures of infrared imaging air-to-air missiles[J]. Infrared and Laser Engineering, 2021, 50(4): 20200240. (in Chinese)
[4]	ZHANG X B. Two-step non-local means method for image denoising[J]. Multidimensional Systems and Signal Processing, 2022, 33(2): 341-366.
[5]	杨良健, 周先春, 崔程程, 等. 基于旋转块的BM3D图像去噪改进算法[J]. 电子测量技术, 2021, 44(22): 108-113.
	YANG L J, ZHOU X C, CUI C C, et al. Improved algorithm of BM3D image denoising based on rotating block[J]. Electronic Measurement Technology, 2021, 44(22): 108-113. (in Chinese)
[6]	SAAD N H, ISA N A M, SALEH H M. Nonlinear exposure intensity based modification histogram equalization for non-uniform illumination image enhancement[J]. IEEE Access, 2021, 9: 93033-93061.
[7]	TAN S F, ISA N A M. Exposure based multi-histogram equalization contrast enhancement for non-uniform illumination images[J]. Ieee Access, 2019, 7: 70842-70861.
[8]	TIAN C W, FEI L K, ZHENG W X, et al. Deep learning on image denoising: an overview[J]. Neural Networks, 2020, 131: 251-275. doi: S0893-6080(20)30266-5 pmid: 32829002
[9]	MURALI V, SUDEEP P V. Image denoising using DnCNN:an exploration study[C]∥ Proceedings of the Advances in Communication Systems and Networks: Select Proceedings of ComNet 2019. Singapore: Springer, 2020: 847-859.
[10]	ZHANG K, ZUO W M, ZHANG L. FFDNet: Toward a fast and flexible solution for CNN-based image denoising[J]. IEEE Transactions on Image Processing, 2018, 27(9): 4608-4622.
[11]	GUO S, YAN Z F, ZHANG K, et al. Toward convolutional blind denoising of real photographs[C]∥Proceedings of the IEEE/CVF conference on computer vision and pattern recognition 2019. Long Beach, CA, US: IEEE, 2019: 1712-1722.
[12]	REN C, HE X H, WANG C C, et al. Adaptive consistency prior based deep network for image denoising[C]∥Proceedings of the IEEE/CVF conference on computer vision and pattern recognition 2021. Nashville, TN, US: IEEE, 2021: 8596-8606.
[13]	ZHANG K, LI Y, LIANG J Y, et al. Practical blind denoising via swin-conv-unet and data synthesis[J]. arXiv preprint arXiv:2203.13278, 2022.
[14]	XIE M, ZHANG Z D, ZHENG W B, et al. Multi-frame star image denoising algorithm based on deep reinforcement learning and mixed poisson-Gaussian likelihood[J]. Sensors, 2020, 20(21): 5983.
[15]	LI Z, WU H, CHENG L L, et al. Infrared and visible fusion imaging via double-layer fusion denoising neural network[J]. Digital Signal Processing, 2022, 123: 103433.
[16]	LIU H M, AN Q, LIU T N, et al. An infrared image denoising model with unidirectional gradient and sparsity constraint on biomedical images[J]. Infrared Physics & Technology, 2022, 126: 104348.
[17]	WEI Z C, WU X J, TONG W, et al. Elimination of stripe artifacts in light sheet fluorescence microscopy using an attention-based residual neural network[J]. Biomedical Optics Express, 2022, 13(3): 1292-1311. doi: 10.1364/BOE.448838 pmid: 35414974
[18]	XU K, ZHAO Y H, LI F Z, et al. Single infrared image stripe removal via deep multi-scale dense connection convolutional neural network[J]. Infrared Physics & Technology, 2022, 121: 104008.
[19]	SOH J W, CHO N I. Deep universal blind image denoising[C]∥Proceedings of the 25th International Conference on Pattern Recognition (ICPR) 2021. Milan, Italy: IEEE, 2021: 747-754.
[20]	ZHOU Z, SIDDIQUEE M M R, TAJBAKHSH N, et al. Unet++: Redesigning skip connections to exploit multiscale features in image segmentation[J]. IEEE transactions on medical imaging, 2019, 39(6): 1856-1867.
[21]	WANG C Y, LIAO H Y M, WU Y H, et al. CSPNet: a new backbone that can enhance learning capability of CNN[C]∥Proceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops 2020. Seattle, WA, US: IEEE, 2020: 390-391.
[22]	ELFWING S, UCHIBE E, DOYA K. Sigmoid-weighted linear units for neural network function approximation in reinforcement learning[J]. Neural Networks, 2018, 107: 3-11. doi: S0893-6080(17)30297-6 pmid: 29395652
[23]	LIU P J, ZHANG H Z, LIAN W, et al. Multi-level wavelet convolutional neural networks[J]. IEEE Access, 2019, 7: 74973-74985. doi: 10.1109/ACCESS.2019.2921451
[24]	HE Z W, CAO Y P, DONG Y F, et al. Single-image-based nonuniformity correction of uncooled long-wave infrared detectors: a deep-learning approach[J]. Applied Optics, 2018, 57(18): D155-D164.
[25]	LIU Q, LI X, HE Z Y, et al. LSOTB-TIR: a large-scale high-diversity thermal infrared object tracking benchmark[C]∥Proceedings of the 28th ACM International Conference on Multimedia 2020. Seattle, WA, US: Association for Computing Machinery, 2020: 3847-3856.
[26]	ZHANG K, LI Y W, ZUO W, et al. Plug-and-play image restoration with deep denoiser prior[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 44(10):6360-6376.