For long-distance and wide field-of-view scenes, infrared target detection has significant challenges owing to the principle of a thermal imager, interference of the atmospheric environment, and attenuation of infrared radiation by long-distance transmission media. Based on the characteristic analysis of small-target infrared images, such as complex background, dim and small targets, low image contrast, and lack of image structures, we reviewed the research status of infrared dim small-target detection from target highlight and background estimation and discussed the development trend of infrared dim small-target detection.

Tandem white OLEDs offer low power consumption, high brightness, and a high color gamut. However, the material and electrical structures of tandem white OLEDs still need to be optimized owing to the outstanding challenges in efficiency, lifetime, and driving voltage. In this study, we focused on the latest research on tandem white OLEDs and summarized the problems in engineering preparation and non-destructive detection method of 3 types of CGLs for high-efficiency tandem white OLEDs. We focused on the latest research on the “all-phosphorescent system,” “harvesting excitons via two parallel channels,” and the “mixed-phosphorescent-TADF system” simultaneously. We summarized the device lifetime problems and discussed structural solutions such as "graded doping" and "four-color mixed-phosphorescent-TADF system."

We proposed a color-corrected underwater illumination image fusion method based on color correction to address uneven color shifts, low contrast, and blurred details in underwater illumination images. First, we exploited the pixel correlation between image channels to compensate for the red channel. Then, based on the color-corrected image, a sharpness-enhanced image is obtained using a nonlinear unsharp masking technique, and a global stretching map is obtained using a restricted histogram with Rayleigh distribution. Finally, we generated the fused image using a multi-scale fusion strategy. The experimental results on a self-built dataset (RULI) showed that the proposed method could remove the inhomogeneous scattering interference of mixed illumination in the imaging process and substantially improve the detail sharpness of the image. The mean values of the image quality assessment metrics UIQM and IE were 4.7399 and 7.7617, respectively, better than those of related algorithms in the existing literature.

Aiming at more redundant background information and low stitching accuracy of the infrared images of the blades taken by UAV (Unmanned Aerial Vehicle), In this study, we proposed a stitching algorithm for infrared wind turbine blade images combining the Chan-Vese model and morphology. First, we subjected the image to median filtering and noise reduction, and a morphological operation improved a level-set algorithm based on the Chan-Vese model to generate the mask of the expression subject. We extracted Harris feature points by removing redundant backgrounds based on the mask. We performed morphological etching on the mask to suppress the pseudo-feature points on the boundary-jagged pixels. We used violent matching and the RANSAC algorithm to screen out effective matching point pairs and calculate the homography matrix to realize matching and splicing. Compared with the Harris stitching algorithm under traditional image segmentation, the stitching accuracy of the improved algorithm significantly improved, and it showed strong robustness in different test scenarios.

Considering the false alarm and real-time requirements of infrared small-target detection under a complex cloud background, a novel algorithm is proposed based on structure tensor screening and local contrast analysis. Combined with the feature that the maximum eigenvalue of the structure tensor of the target area is larger than that of other background areas, the proposed algorithm can filter out most nontarget areas and retain a few suspicious areas. Local contrast calculation performed on suspicious areas can enhance the target, suppress the residual background, and effectively reduce computation. The algorithm steps are as follows: first, we constructed the structure tensor matrix within the local image area captured by the sliding window, and where the maximum eigenvalue is larger than the threshold is marked as a suspicious area. Then, we calculated the ratio-difference joint local contrast. Finally, we adopted an adaptive threshold segmentation on the saliency map to extract the real target. Experimental results showed that the proposed algorithm can achieve a higher detection rate, lower false alarm rate, and shorter running time under a complex cloud background.

We proposed a multimodal crowd counting algorithm based on RGB-Thermal (RGB-T) images (two-stream residual expansion network) in crowd counting, given scale changes, uneven pedestrian distribution, and poor imaging conditions at night. It has a front-end feature extraction network, multi-scale residual dilation convolution, and global attention modules. We used the front-end network to extract RGB and thermal features, and the dilated convolution module further extracted pedestrian feature information at different scales and used the global attention module to establish dependencies between global features. We also introduced a new multi-scale dissimilarity loss method to improve the counting performance of the network and conducted comparative experiments on the RGBT crowd counting (RGBT-CC) and DroneRGBT datasets to evaluate the method. Experimental results showed that compared with the cross-modal collaborative representation learning (CMCRL) algorithm on the RGBT-CC dataset, the grid average mean absolute error (GAME (0)) and root mean squared error (RMSE) of this algorithm are reduced by 0.8 and 3.49, respectively. On the DroneRGBT dataset, the algorithm are reduced by 0.34 and 0.17, respectively, compared to the multimodal crowd counting network (MMCCN) algorithm, indicating better counting performance.

We proposed a new object detection method based on the CSE-YOLOv5 (CBAM-SPPF-EIoU-YOLOv5) model for insufficient multi-scale feature learning ability and the difficulty of balancing detection accuracy and model parameter quantity in remote sensing image object detection algorithms in complex task scenarios. We built this method on the YOLOv5 model's backbone network framework and introduced a convolutional attention mechanism layer into the shallow layers to enhance the model's ability to extract refined features and suppress redundant information interference. In the deep layers, we constructed a spatial pyramid pooling fast (SPPF) with a tandem construction module and improved the statistical pooling method to fuse multi-scale key feature information from shallow to deep. In addition, we further enhanced the multi-scale feature learning ability by optimizing the anchor box mechanism and improving the loss function. The experimental results demonstrated the superior performance of the CSE-YOLOv5 series models on the publicly available datasets RSOD, DIOR, and DOTA. The average mean precisions (mAP@0.5) were 96.8%, 92.0%, and 71.0% for RSOD, DIOR, and DOTA, respectively. Furthermore, the average mAP@0.5:0.95 at a wider IoU range of 0.5 to 0.95 achieved 87.0%, 78.5%, and 61.9% on the same datasets. The inference speed of the model satisfied the real-time requirements. Compared to the YOLOv5 series models, the CSE-YOLOv5 model exhibited significant performance enhancements and surpassed other mainstream models in object detection.

The segmentation accuracy of substation equipment in infrared images captured by a UAV directly affects the results of thermal fault diagnosis. We proposed a multimodal path aggregation network (MPAN) that fuses visible and infrared images to address the problem of low segmentation accuracy of substation equipment in complex infrared backgrounds. First, we extracted and fused the features of two modal images, and considering the differences in the feature space of the two modal images, we proposed the adaptive feature fuse module (AFFM) to fuse the two modal features fully. We added a bottom-up pyramid network to the backbone with multi-scale features and a laterally connected path enhancement. Finally, we used dice coefficients to optimize the mask loss function. The experimental results showed that the fusion of multimodal images can enhance the segmentation performance and verify the effectiveness of the proposed modules, which can significantly improve the accuracy of the segmentation of substation equipment instances in infrared images.

We proposed a parameter self-tuning bi-histogram equalization method to solve saturation and detail loss in infrared image enhancement. We decomposed an input image into two independent sub-images according to the golden ratio of the gray cumulative probability density and modified each sub-image histogram through a multi-scale adaptive weighing process with input image exposure and sub-image gray-level interval information. Subsequently, we performed the equalization of the two corrected sub-histograms independently and combined the two equalized sub-images into a single output image. A test on 100 infrared images in a public dataset-INFRARED100 showed that, compared with brightness preserving bi-histogram equalization (BBHE), bi-histogram equalization with a plateau limit (BHEPL), and exposure-based sub-image histogram equalization (ESIHE), the images enhanced by the proposed method have appropriate contrast and greater average information entropy. We increased the peak signal-to-noise ratio (PSNR), structural similarity (SSIM) index, and absolute mean brightness error (AMBE) by at least 17.2%, 4.0%, and 56.2% on average. The experiments illustrated that the proposed method is adaptable to infrared images with different brightness characteristics, effectively improving the contrast between the infrared image object and background. This method is superior to noise suppression, brightness, and detail preservation methods.

We proposed a two-scale image-fusion method for infrared and visible light image fusion based on guided filtering to reduce the complexity of multi-scale decomposition fusion algorithms and improve the adaptability of fused images to human visual characteristics. First, we used guided filtering to enhance the visible image and decomposed the source images into base and detail layers using guided filtering. In the fusion rules of the detail layer, we adopted the energy protection methods and detail extraction. Finally, we combined the fused detail layer with the base layer to synthesize the fusion results. The experimental results showed that the proposed method improves the visual effect, detail processing, and edge protection. We discussed the impact of visible image enhancement on fusion methods from experimental data. Enhancement can improve the fusion effect, but the fusion method is key in image fusion.

In this study, we proposed an infrared and visible image fusion algorithm that combines PIE and CGAN to make unmanned agricultural machinery perceive environmental information promptly and avoid accidents during production in complex environments. First, we trained the CGAN using an infrared image and corresponding saliency regions. The infrared image is input into the trained network to obtain the saliency region mask. After morphological optimization, we performed image fusion based on the PIE. Finally, we enhanced the fusion results by contrast processing. This algorithm can realize fast image fusion and satisfy the requirements for real-time environmental perception of unmanned agricultural machines. In addition, the algorithm retains the details of visible images and highlights important information concerning humans and animals in infrared images. It performs well in standard deviation and information entropy.

Telephoto common optical path imaging components are widely used in photoelectric reconnaissance pods, and the technical development of telephoto common optical path fast mirrors for composite axis image stabilization has become an inevitable trend. This study introduced the main components of telephoto common optical path imaging components. We realized the composite axis control and flyback compensation control strategy based on a fast mirror and analyzed and calculated its working timing and key parameters. We developed a fast mirror based on a telephoto common optical path imaging device, and simultaneously realized secondary image stabilization and flyback compensation within one frame of the image. We improved the reconnaissance range, image stabilization accuracy, and the search effect of medium- and high-altitude photoelectric reconnaissance pods.

We can use the infrared radiation characteristics of a target for target recognition. Data on infrared radiation characteristics obtained by out-field infrared imaging equipment is significant in evaluating early warning, reconnaissance, and stealth effects. It is difficult to obtain the response coefficients of out-field infrared imaging equipment. We introduced and compared radiometric calibration methods using a collimator and an extended-area blackbody. We conducted experiments using different calibration methods and then provided response coefficients of the out-field infrared imaging equipment. The long-distance radiometric calibration results showed different response coefficients at different distances. An infrared imaging system conducted calibration experiments with different working times and fusions. The radiometric out-of-focus calibration results showed that diffusion is not the main factor influencing calibration. Calibration experiments for different working times also showed that the response coefficients remained unchanged. The factors affecting the radiometric calibration of the out-field infrared imaging equipment are environmental radiation, path radiation, and path transmission. Short-distance radiometric calibration using an extended-area blackbody is necessary to obtain the response coefficients of the out-field infrared imaging equipment. If the radiometric calibration distance is less than 10 m, the error between the short-and long-distance radiometric calibrations is approximately 5%. This research helps out-field radiometric calibration of ground-based infrared imaging equipment and designs a radiometric calibration–measuring system.

In this study, the surface processing of cadmium zinc telluride (CZT) substrates was studied, which revealed surface dislocation defects. The surface processing mechanism and influence of the process parameters on the surface of the CZT substrates, including mechanical grinding, mechanical polishing, chemical mechanical polishing, and chemical polishing, are presented. Moreover, three types of chemical etchants, Everson, Nakagawa, and EAg, which reveal dislocation defects on the surface of CdZnTe with different crystal orientations, were also investigated.

Aiming at the problems of low recognition accuracy and slow detection speed of existing metal oxide arrester (MOA) infrared image fault detection methods, a MOA infrared image fault detection method based on improved YOLOv3 is proposed. Firstly, darknet19 network is used to replace the original darknet53 network of YOLOv3. During feature learning, the target frames in MOA images are analyzed by K-means clustering algorithm according to different MOA length width ratios in samples. The anchor frames in the center of samples are re clustered to get the appropriate number and size of anchor frames. Finally, the improved YOLOv3 model is used to complete the MOA infrared image fault detection. The experimental results show that the recognition accuracy of the improved model reaches 96.3%, and the recognition speed is 6.75ms.