Current Issue

2023, Volume 45,  Issue 9

Image Processing and Simulation
Infrared and Visible Image Fusion Based on N-RGAN Model
SHEN Yu, LIANG Li, WANG Hailong, YAN Yuan, LIU Guanghui, SONG Jing
2023, 45(9): 897-906.
Abstract(30) HTML (7) PDF(16)
At present, infrared and visible image fusion algorithms still have problems such as low applicability to complex scenes, large loss of detail and texture information in fusion images, and low contrast and sharpness of fusion images. In view of the above problems, this study proposes an N-RGAN model that combines a non-subsampled shearlet transform (NSST) and a residual network (ResNet). Infrared and visible images are decomposed into high- and low-frequency sub-bands using NSST. The high-frequency sub-bands are spliced and input into the generator improved by the residual module, and the source infrared image is taken as the decision standard to improve network fusion performance, fusion image detail description, and target-highlighting ability. The salient features of infrared and visible images are extracted, and the low-frequency sub-bands are fused by adaptive weighting to improve image contrast and sharpness. The fusion results of the high- and low-frequency sub-bands are obtained by the NSST inverse transformation. Based on a comparison of various fusion algorithms, the proposed method improves peak signal-to-noise ratio (PSNR), average gradient (AVG), image entropy (IE), spatial frequency (SF), edge strength (ES), and image clarity (IC), thereby improving infrared and visible light image fusion effects in complex scenes, alleviating information loss in image detail texture, and enhancing image contrast and resolution.
A Visible and Infrared Image Fusion Algorithm Based on Adaptive Enhancement and Saliency Detection
CHEN Sijing, FU Zhitao, LI Ziqian, NIE Han, SONG Jiawen
2023, 45(9): 907-914.
Abstract(10) HTML (0) PDF(9)
This paper proposes a visible and infrared image fusion algorithm to solve the problem of the poor visibility of visible images and control the input volume of visible and infrared images. The proposed method combines image adaptive enhancement with uniqueness (U), focus (F), and object (O) saliency detection. First, an adaptive enhancement algorithm was applied to the visible image to improve the visibility of the textural details and normalize the infrared image. Second, the processed image was decomposed into a detail layer and base layer using guided filtering. A weight map of the detail layer was generated using saliency detection to improve the accuracy of the fusion of the background information of the visible image and the edge information of the infrared image in the detail layer. Finally, the fused image was obtained by combining the detail and base layers. To verify the performance of the proposed algorithm, five fusion evaluation indices: image entropy, average gradient, edge intensity, spatial frequency, and visual fidelity, were selected to quantitatively analyze the fused images. The YOLO v5 network was used to perform target detection for each fusion algorithm. The results show that the proposed algorithm achieved the optimal average accuracy in terms of the qualitative, quantitative, and target detection evaluation indexes of fusion.
Infrared and Visible Image Fusion Based on Guided Filter and Sparse Representation in NSST Domain
WU Lingxiao, KANG Jiayin, JI Yunxiang
2023, 45(9): 915-924.
Abstract(9) HTML (1) PDF(10)
Image fusion technology aims to solve the problem of insufficient and incomplete information provided by a single-modality image. This paper proposes a novel method based on guided filter (GF) and sparse representation (SR) in the non-subsampled shearlet transform (NSST) domain, to fuse infrared and visible images. Specifically, ① the infrared and visible images are respectively decomposed using NSST to obtain the corresponding high-frequency and low-frequency sub-band images; ② The GF-weighted fusion strategy is exploited to fuse the high-frequency sub-band images; ③ Rolling guidance filter (RGF) is used to further decompose the low-frequency sub-band images into base and detail layers, whereby the base layers are fused via SR, and the detail layers are fused using local maximum strategy which is based on consistency verification; ④ An inverse NSST is performed on the fused high-frequency and low-frequency sub-band images to obtain the final fusion result. Compared to those of other methods, experimental results on public datasets show that the fusion result obtained by the proposed method has richer texture detail and better subjective visual effects. In addition, the proposed method achieves overall better performance in terms of objective metrics that are commonly used for evaluating fusion results.
A Small Target Detection Algorithm from UAV Perspective
LI Yang, WU Lianquan, YANG Haitao, NIU Jinlin, CHU Xianteng, WANG Huapeng, ZOU Qinglong
2023, 45(9): 925-931.
Abstract(12) HTML (8) PDF(5)
The use of unmanned aerial vehicles (UAVs) for effective real-time monitoring of small targets, such as people, cars, and objects in the scene area, can help maintain public security. To address the problems of small-target occlusion, overlapping, and interference of complex environments in UAV images, a small-target detection algorithm is proposed from the UAV perspective. The algorithm uses the YOLOX network as the baseline system. First, the neck part of the network increases the output feature graph to reduce the receptive field, thereby improving the performance of the network details, and the detection head of the small-sized feature graph is deleted to improve the detection rate of small targets. Second, the anchor-free association mechanism is used to reduce the influence of noise in the truth tag while simultaneously reducing the parameter setting to speed up network operations. Finally, a true proportion coefficient is proposed for small targets to calculate position loss, thereby increasing the penalty for misjudging small targets, which makes the network more sensitive to small targets. Experiments on the VisDrone2021 dataset using this algorithm showed that the mAP value increased by 4.56%; the number of parameters decreased by 29.4%; the amount of computation decreased by 32.5%; and the detection speed increased by 19.7% compared with those of the baseline system, which is an advantage over other mainstream algorithms.
Image Enhancement Algorithm Based on Texture Prior and Color Clustering
LIU Zhengnan, LIU Chunjing
2023, 45(9): 932-940.
Abstract(4) HTML (2) PDF(4)
To address the problems of traditional multiscale retinex with color restoration (MSRCR), such as texture information weakening, partial information loss, and poor enhancement effects, an image enhancement algorithm based on texture priors and color clustering is proposed. First, prior to image enhancement, texture information is extracted for further processing. Second, considering the uneven illumination distribution, a color-clustering algorithm is proposed for image segmentation enhancement. In addition, for logarithmic domain mapping, a mapping scheme based on mean square value and mean square error is proposed based on block processing. Finally, in evaluating the enhancement algorithm, information entropy and natural statistics of the image are used to evaluate the effectiveness of the enhanced image. The experimental results show that the average entropy of the proposed method reached 7.4934 and the average of natural statistical properties reached 4.0903. The algorithm effectively enhances the details of the image, makes the image more natural, and further improves image quality.
Fusion Algorithm of Infrared and Visible Images Based on Semantic Loss
DING Huabin, DING Qiwen
2023, 45(9): 941-947.
Abstract(8) HTML (2) PDF(8)
In this study, we propose an infrared and visible image fusion algorithm based on semantic loss, to ensure that the generated images contain more semantic information through semantic loss, thereby satisfying the requirements of advanced vision tasks. First, a pre-trained segmentation network is used to segment the fused image, with the segmentation result and label map determining the semantic loss. Under the joint guidance of semantic and content losses, we force the fusion network to guarantee the quality of the fused image by considering the amount of semantic information in the image, to ensure that the fused image meets the requirements of advanced vision tasks. In addition, a new feature extraction module is designed in this study to achieve feature reuse through a residual dense connection to improve detail description capability while further reducing the fusion framework, which improves the time efficiency of image fusion. The experimental results show that the proposed algorithm outperforms existing fusion algorithms in terms of subjective visual effects and quantitative metrics and that the fused images contain richer semantic information.
Infrared Image Stitching Method for Sandwich Bulkhead Structure
SHENG Tao, ZHENG Jinhua, XIANG Ping, CHEN Chao, XU Hongjie, JIANG Haijun
2023, 45(9): 948-953.
Abstract(4) HTML (1) PDF(5)
A sandwich bulkhead structure has a large volume. Because infrared non-destructive testing technology can only detect a small area, hundreds of tests are required to detect the sandwich bottom structure, which is not conducive to determining the location of defects in a single project. This article proposes a fixed overlapping-area infrared image stitching method in which the overlapping area is fused using weighted fusion to effectively eliminate seams. The six sub-regions on the side of the sandwich bottom structure are stitched using an arc-shaped method. This approach not only solves the problem of stitching in multiple projects but also greatly improves the accuracy and efficiency of defect detection. After collecting and annotating the defects in the project, the stitched image directly displays the location information of the defects, which greatly facilitates the identification of defect information in the sandwich bulkhead structure.
Shortwave Infrared Image Dehazing Based on Dark Channel Prior
LIU Yanqing, LI Zhongwen, YU Shikong, LIU Yunyi, YAO Wenting, GE Zhihao, JI Li, ZHANG Baohui
2023, 45(9): 954-961.
Abstract(8) HTML (1) PDF(5)
To solve the problems of blurred image quality and low-resolution weather haze in shortwave infrared imaging systems, a shortwave infrared image-defogging algorithm based on a dark channel prior is proposed. First, the algorithm obtains the dark-channel image data using an improved dark-channel prior. Then, the atmospheric light is estimated based on the dark channel data. To avoid local highlights or blurred details of the target, the transmittance map is refined and enhanced using guided filtering and multi-scale retinex (MSR). Finally, the defogged image is inverted using the atmospheric scattering model. The shortwave infrared image processed by this algorithm was verified in terms of subjective vision and objective indicators, displaying a remarkable defogging effect, rich details, and appropriate brightness.
Underwater Evidence Detection Method Based on Polarization Fusion Image
GAO Yi, YU Jinqiang, ZHANG Xiaodong, DUAN Jin
2023, 45(9): 962-968.
Abstract(4) HTML (6) PDF(6)
Polarization detection technology can highlight targets in complex background environments, providing us with clearer and more accurate target recognition. However, research on the use of polarization imaging in courtroom science, regarding the detection and search for underwater evidence, is lacking. To address this issue, this study fuses the polarization and target intensity images using a polarization imaging device. After decomposing the images using non-subsampled shear waves (NSST) into low and high-frequency sub-bands, a simplified impulse-coupled neural network model with adaptive parameters is proposed for the high-frequency sub-band, and an adaptive weighting fusion rule, based on region energy, is used for the low-frequency sub-band. Correlation algorithm comparison experiments were conducted for three typical targets at visible wavelengths. The experimental results show that underwater evidence can be effectively detected using polarization imaging technology. The image fusion algorithm proposed in this paper effectively highlights the detailed features of underwater evidence, verifying the effectiveness of polarization detection technology for underwater evidence imaging, which is conducive to breaking through the existing research gap in the field of courtroom science.
System & Design
Determination of Optical Constants of Transparent Solids Based on Double Thickness Transmittance Model of Polynomial Root
YANG Baiyu, WU Xiaoliang, WANG Cuixiang, WANG Weiyu, LI Lei, FAN Qi, LIU Jing, XU Cuilian
2023, 45(9): 969-973.
Abstract(3) HTML (1) PDF(5)
In determining the optical constants of transparent solids using spectral inversion methods, certain problems such as inversion errors and computational time consumption, have to be solved. This study establishes two spectral transmittance equations with the thickness satisfying the integer ratio based on the traditional double-thickness transmittance model. A polynomial equation related to the extinction coefficient is obtained through an algebraic operation, and the extinction coefficient is calculated by solving and selecting a real root greater than 0 and less than 1. Subsequently, the unitary quadratic equation is solved for interface reflectivity, thereby selecting the roots that are greater than 0 and less than 1 to calculate the refractive index. In the process of determining the optical constants, the new method does not suffer from inversion errors, time-consuming iterative calculations, or multivalue problems. As an application example, the optical constants of CaF2 and Si were calculated using the experimental data of double-thickness transmittance in the literature, and the results were compared with those in the literature. The results show that the new method is superior to traditional spectral inversion methods and provides a new option for high-precision determination of optical constants of transparent solids.
Nondestructive Testing
Infrared Thermal Imaging Defect Detection of Photovoltaic Module Based on Improved YOLO v5 Algorithm
KONG Songtao, XU Zhenze, LIN Xingyu, ZHANG Chunqiu, JIANG Guoqing, ZHANG Chunqing, WANG Kun
2023, 45(9): 974-981.
Abstract(10) HTML (0) PDF(7)
To solve the problem of difficulty in extracting features and poor real-time performance of existing photovoltaic power station defect identification methods, which lead to low identification accuracy of photovoltaic module defect detection, this paper proposes a photovoltaic power station infrared thermal imaging defect detection method based on an improved YOLO v5 algorithm. The improved YOLO v5 algorithm primarily adds an attention mechanism SE module to the original core and improves the loss function from GIoU to EIoU to enhance the model convergence effect. Finally, the knowledge graph (KG) module is used to balance the feature pyramid structure and optimize the model to improve the YOLO v5 algorithm's recognition accuracy and convergence effects. The improved network structure was applied to the YOLO v5s model, whereby the average detection accuracy mAP used in the detection of infrared images of photovoltaic power plants reached 92.8%, which is 4.5% higher than that of the original YOLO v5s algorithm (88.3%). The effect of convergence on the precision and recall rate was also improved compared with the original YOLO v5 algorithm model. By applying the enhanced network structure to the three models (l, m, and x), detection accuracy was also improved. Consequently, the improved YOLO v5 algorithm is suitable for the four models.
Detection and Recognition of Metal Fatigue Cracks by Bi-LSTM Based on Eddy Current Pulsed Thermography
LIN Li, JIANG Jing, ZHU Junzhen, FENG Fuzhou
2023, 45(9): 982-989.
Abstract(6) HTML (0) PDF(5)
Eddy current pulsed thermography is a new nondestructive testing method that is widely used in metal structure testing. However, the extraction of features for crack detection and identification relies on manual experience, and the degree of automation and intelligence is insufficient. By combining the characteristics of eddy current pulsed thermography with a recurrent neural network (RNN), a bidirectional long short-term memory (Bi-LSTM)-based eddy current pulse thermography method is proposed for metal fatigue crack classification and recognition. The Bi-LSTM model was designed to enhance the transient information in the feature vectors. In the experiments, an eddy current heating device was used to heat the tested metal specimens. A real-time dataset was created using an infrared thermal camera that collected sequences of images. The Bi-LSTM model was trained on thermal images of cracks of different sizes and tested. Experimental analyses show that the Bi-LSTM network can be effectively applied for metal fatigue crack detection and recognition, with the detection accuracy reaching 100% for the cracks used in the experiments, which is superior to that of traditional neural networks and other deep learning models.
Ir Applications
Vehicle Infrared Illumination System for Assistant Driving
ZHUANG Yabao, LIU Jie, XUE Hao, ZHU Xiangbing
2023, 45(9): 990-995.
Abstract(5) HTML (0) PDF(4)
An onboard infrared supplementary illumination system is designed through software simulations to solve the problem of low recognition accuracy caused by insufficient ambient illumination and rapid changes in ambient light intensity in an advanced driving assistance image recognition system. IR light-emitting diodes (LEDs) are used as light sources. After collimating the light emitted by the light source with a paraboloid mirror, a compound eye lens is used to realize a rectangular spot with uniformity greater than 90% 25 m away from the light source with radiation energy-use efficiency reaching 98%. The target surface is illuminated using 30 light-source module arrays, and the average irradiance of the target surface is greater than 0.8 W/m2. The results show that the assembly tolerance of a single-compound eye lens is looser, and the design method is flexible and suitable for use in other lighting systems.
Energy Consumption Analysis of Building Window Defects Based on Infrared Image Processing
ZHANG Lingling, ZHANG Jiran, XU Ao, REN Panpan, DING Libin
2023, 45(9): 996-1004.
Abstract(15) HTML (1) PDF(8)
The differential pressure method, which combines infrared thermal imaging and image processing technologies, is used to detect air infiltration of building exterior windows. Infrared images of the exterior windows of the building were collected using an infrared thermal imager and then processed using infrared image processing technology. Exterior window defects were detected from abnormal areas in the infrared images, and the area of the defects was calculated to establish an infrared detection model for exterior window defects. Based on the indoor and outdoor temperature difference, defect area of the outer window, and air infiltration amount measured in the experiment, a calculation model was established for the amount of air infiltration for the building's outer window. The model was combined with the infrared detection model for building window defects, to quantitatively analyze the energy consumption caused by the defects. The results show that the maintenance of exterior window defects can reduce energy consumption of the exterior window and improve energy savings. For every 1 cm2 reduction in the air infiltration area of exterior windows, 66146 kJ of energy can be saved annually. For each level of airtightness improvement of exterior windows, 110012 kJ of energy per unit area of exterior windows can be saved annually,