摘要:Organic semiconductor materials exhibit promising prospects for flexible device applications due to their excellent solution processability and superior charge transport performance. This work proposed a method for preparing a stretchable organic semiconductor material. By using benzophenone as a photoinitiator, the conjugated polymer 3,6-dithiophene-2-yl-2,5-dihydropyrrolo [3,4-c] pyrrole-1,4-diketone (4Si) was crosslinked with polydimethylsiloxane. The alterations in the electrical and mechanical properties of the polymer films were thoroughly investigated on both pre- and post-crosslinking. Compared to the uncrosslinked 4Si and blend films, the electrical properties of the crosslinked films slightly decreased, while the mechanical properties were enhanced. The crack onset-strain has increased from 30% to 50%. Furthermore, at 100% strain, the crosslinked film exhibited a mobility of 0.36 cm2·V-1·s-1 in the parallel strain direction, which is 1.44 times higher than the initial mobility.
摘要:Perovskite quantum dot (QD)light-emitting diodes are novel generation display technique with great potential.All-inorganic CsPbX3 (X=Cl,Br,I)QDs exhibit excellent optoelectronic properties. However, the synthesis of CsPbX3 QDs usually requires the long-chain ligands such as oleic acid (OA) and oleylamine (OAm). The dynamic binding between these ligands and QDs makes them easily detach from the QD surface, thereby leading to the poor stability. In addition, the insulating properties of OA and OAm are not conducive to the charge transfer. In this work, we adopt the strategy of replacing OA with 2-hexyldecanoic acid (DA) partially, systematically studying the influence of DA/OA molar ratios on the optical properties of CsPbBr3 QDs. The results show that CsPbBr3 QDs modified by DA exhibit a high photoluminescence quantum yield of 88.64%. Based on this strategy, the all-solution-processed light-emitting diodes achieve the luminance of 155 cd·m-2 and external quantum efficiency (EQE) of 0.14%. Subsequently, by optimizing the energy levels between the perovskite layer and the hole transport layer, the luminance and EQE are further improved to 436 cd·m-2 and 0.20%.
摘要:Depth compression processing for three-dimensional display content is an effective solution to the problem of insufficient depth range in 3D raster displays. However, commonly used compression methods inevitably cause geometric deformation in the main subjects of the scene. This paper proposes a depth optimization method for dense viewpoint3D raster displays based on 3D point clouds. By reconstructing the 3D point cloud from the stereo disparity map of the 3D scene, the main subject’s point cloud is segmented, and depth position adjustments are made only for subjects exceeding the display’s depth range, thereby maintaining the geometric structure of the scene’s main subjects. This achieves overall depth compression of the 3D scene without altering the geometric integrity of the main subjects. Through a series of statistical experiments focused on the visual experience, the proposed method received an approval rating exceeding 77% from each participant and surpassed 80% for every experimental scenario. The results affirm the method’s significant enhancement of the audience’s subjective perception, compressing the overall depth of the scene while ensuring no deformation of the main subjects, thus providing viewers with a natural viewing experience and strong depth perception of the scene’s main subjects.
关键词:autostereoscopic 3D display;3D point cloud;instance segmentation;depth compression
摘要:Holographic optical element (HOE) is one of the important devices for holographic 3D display. In order to solve the problem of parallax mismatch when HOE is used for binocular holographic display, this paper proposes and implements a holographic display binocular parallax construction method based on HOE. At the same time, the aberrations of the components are analyzed and corrected. By analyzing the production principle and imaging rule of HOE, combining the virtual model rendered by ray tracing technology and the real scene model, a holographic binocular display system based on HOE is designed, and a binocular parallax construction method is proposed, that is, a real-time view image containing binocular parallax relationship is provided for the observer when acquiring the coordinate position of the human eye. By further analyzing the cause and process of image distortion in holographic display system, an optimization algorithm to remove the distortion is proposed. Experiment results show that the proposed method can not only realize the HOE display holographic 3D image in real time, but also solve the problem of parallax mismatch and image distortion in the holographic binocular display of HOE.
摘要:Holographic retinal projection display(RPD) is one of the near eye displays that can alleviate vergence accommodation conflict(VAC) due to its large depth of field. However, an excessive depth of field (DOF) will cause a loss of depth cues whose image clarity does not vary with depth. To solve this problem, a band-limited random phase based holographic RPD is proposed by using the band-limited random phase to control the light beam width into the viewer’s pupil. And the depth of field will be limited by the controlled beam width. Thus, the depth cue of defocus can be enhanced. The process of the band-limited random phase based RPD can be divided into three steps: Firstly, the band-limited random phase is calculated by the light beam width into the pupil and the depth of the target image. Secondly, the target image is multiplied by the spherical wave phase and the band-limited random phase. The hologram is obtained by a one-step Fresnel diffraction from the processed complex amplitude distribution in the target image plane. Lastly, the spatial light modulator loaded with the hologram is illuminated by the collimated laser light source. And the reconstructed object light converges on the pupil which achieves the depth cue enhanced RPD. The experiment verifies that the band-limited random phase based holographic RPD has the advantages of low scattering and high reproduction quality. And it can flexibly control the DOF of the target image and provide the observer with depth cues of focus blur.
关键词:3D display;holographic display;Retinal projection display;Band-limited random phase
摘要:In computational holography, the realization of multi-plane holographic display based on layer-based method is one of the most commonly used methods. Aiming at the problem of crosstalk between multiple planes in layer-based method, this paper combines the advantages of random phase and quadratic phase, and proposes a multiplanar holographic display method based on hybrid phase. First, a suitable weighting factor is selected to generate the hybrid phase instead of the random phase as the initial phase. After iteration using the iterative Fourier transform algorithm, each obtained sub-hologram is superimposed with Fresnel zone plate with different reconstruction distances. Then, each hologram is superimposed with complex amplitude, and a phase-only hologram is obtained after the phase is obtained, thus the layered image is reconstructed on the specified plane. A multi-plane holographic display system is constructed using a liquid crystal on silicon to carry out relevant optical experiments. The experimental results show that the structural similarity parameter of this method is improved (up to 4.6%) compared with the random phase-based method, which attenuates the crosstalk between multiple holographic reconstruction planes and improves the reproduction quality of the multi-plane holographic display.
关键词:multi-plane display;layer-based method;hybrid phase;phase-only hologram;the fresnel zone plate
摘要:Pointing accuracy is one of the core indicators in the field of beam control, and the level of pointing accuracy directly determines the performance of the beam controller. In order to solve the problems of large beam deflection angle error and deterioration of pointing accuracy caused by the limitation of liquid crystal phased array fabrication process, a beam pointing accuracy optimization algorithm based on improved Harris hawk optimization algorithm (HHO) is proposed. First, the influencing factors are analyzed, reasonable device parameters are selected. Then, the Harris hawk algorithm (HHO) is used to optimize the beam pointing accuracy of the liquid crystal phased array. Finally, in view of the difficulty of the Harris hawk algorithm being easily trapped in the local optimal value, this paper designs the Harris hawk algorithm that takes into account the overall and local optimization strategies (G&L-HHO). In terms of the global optimization strategy, the Cauchy distribution function is selected to increase the initial population diversity and improve the algorithm’s global search efficiency. In terms of local optimization strategies,the optimization performance of the Harris hawk algorithm is enhanced through adaptive cosine weighting factors. The comparative results of simulation experiments show that the G&L-HHO algorithm proposed in this study optimizes the normalization accuracy error from orders of magnitude to orders of magnitude, and greatly reduces the fluctuation amplitude of normalization accuracy error. This study significantly improves the accuracy of beam pointing and has advantages of high convergence accuracy, fast convergence speed, and strong robustness.
摘要:During the camera imaging process, a gradual halo effect may occur due to changes in the viewing angle, resulting in a phenomenon of bright in the middle and dark around the image. The presence of gradual halo results in the loss of some edge texture information in the image, greatly affecting the performance of machine vision processing. To address this issue, this article aims to improve the Retinex-Net network model by correcting image clarity and improving denoising performance. Firstly, in order to maintain the high resolution of the corrected image while improving the receptive field, this paper adds dilated convolution on the basis of the original network model. Secondly, the algorithm improves the denoising method to a dense residual network denoising method, with the aim of densely extracting each layer's features of the vignetting image, preserving more of the image's detailed characteristics and suppressing noise. Finally, this article constructs a dataset of vignetting images and verifies the correction performance of the proposed vignetting correction algorithm on the test set. Compared with the original network model before improvement, the algorithm in this paper improves by 0.293 in SSIM value, 0.727 in PSNR value, and 0.095 in RMSE value. Compared with correction algorithms such as minimizing image entropy, adaptive compensation Retinex, and radial gradient symmetry, the algorithm in this paper has better correction performance and is more suitable for observation and understanding visually.
摘要:The optical flow method cannot fully exploit the facial color information of micro-expressions, resulting in low recognition accuracy. Therefore, this paper proposes a multi-attention dual-flow network method based on color and optical flow. Firstly, the facial color difference maps are obtained in the CIE Luv color space, and the emotional-physiological features are extracted to compensate for the singularity and limitation of the micro-expression optical flow features. Then, the PAM module and ECA block are combined in parallel to obtain the lightweight dual-attention module, which extracts the spatial and channel key features. Finally, a cross-attention mechanism is designed to obtain mixed features of color and optical flow. The mixed features are fused with spatial channel key features for micro-expression classification. The model is evaluated experimentally using leave-one-out cross-validation. The accuracy and F1 scores reach 69.18% and 67.04% on the SAMM dataset, and 72.38% and 70.85% on the CASME Ⅱ dataset. The experimental results are superior to the current mainstream algorithms, further proving the effectiveness of the proposed model and its modules in micro-expression recognition.
摘要:A single image super-resolution reconstruction method for splitting attention networks is proposed to address the problems of lack of texture details, insufficient feature extraction, and unstable training in the existing generation of adversarial networks under large-scale factors. Firstly, the generator is constructed using the split attention residual module as the basic residual block, which improves the generator's feature extraction ability. Secondly, Charbonnier loss function with better robustness and focal frequency loss are introduced into the loss function to replace the mean square error loss function, and regularization loss smoothing training results are added to prevent the image from being too pixelated. Finally, spectral normalization is used in both the generator and discriminator to improve the stability of the network. Compared with other methods tested on Set5, Set14, Urban100 and BSDS100 test sets at a magnification factor of 4, the peak signal-to-noise ratio of this method is 1.419 dB higher than the average value of other comparison methods in this article, and the structural similarity is 0.051 higher than the average value. Experimental data and renderings indicate that this method subjectively has rich details and better visual effects, while objectively has high peak signal-to-noise ratio and structural similarity values.
摘要:For some existing registration methods still suffer from poor accuracy and low efficiency in low overlap conditions, a registration method based on reliable optimal transport is proposed. Firstly, the key points and their feature information are used to form point pairs. The sample consensus algorithm is adopted to reject the wrong point pairs and complete the coarse registration. The initial reliable points are identified while optimizing original position. Secondly, in the process of solving the optimal transport for fine registration, the reliable points involved in the transmission are dynamically adjusted according to the iteration of transport plan and update strategy, which guarantees efficiency and reliability of the registration. To verify the effectiveness of the proposed method, some models in the Stanford standard graphics library and 3DMatch dataset are selected as registration objects, and the proposed method is compared with three common types of registration methods. Experiments results prove that the proposed method improves the accuracy by more than 30% and reduces the running time by more than 25%, which can still maintain excellent registration results in the case of several types of models and various missing conditions.
关键词:point cloud registration;low overlap;reliable points;optimal transport
摘要:Low-light images suffer from low contrast and low signal-to-noise ratio, and the low quality of captured images seriously affects the subsequent observation and measurement. To enhance the quality of low-light images in complex application scenarios, this paper designs an iterative propagation network to accomplish the image enhancement task of micro-optical images in a fast and robust manner. Firstly, this work designs a multi-stage prediction model to model the illumination prediction task incrementally and enhances the nonlinear fitting capability of the model, thus adapting to unknown realistic situations. Considering the inference burden caused by the multi-stage model, this work also constructs iterative loops based on the Retinex forward propagation model to ensure each stage of the multi-stage model converges to a similar or even the same state to optimize the inference process, and significantly improves the inference speed while enhancing the model performance. This work conducts comparative experiments based on publicly available datasets, the average values of peak signal-to-noise ratio and structural similarity are 11.8% and 3.5% higher than the previous best comparison algorithms, respectively. The iterative propagation network is also used to enhance real-world low-exposure photographic images and low-excitation fluorescence microscopy images, and the experimental results demonstrate its excellent image enhancement performance and generalization.
关键词:neural network;Low-light image enhancement;Cascaded structure;retinex model
摘要:To further enhance the segmentation accuracy of deep learning semantic segmentation method on complex street images, this paper proposes a semantic segmentation network (MDFNet) incorporating multi-dimensional features based on PointRend network of street image. Firstly, the algorithm builds a target area enhancement module to optimize the feature extraction sub-network, which self-adaptively refines the intermediate feature map in each convolutional block of the deep network. Thus, the module enhances the fine extraction of multi-dimensional feature information of complex street images. Secondly, the paper introduces feature pyramid grid during feature fusion. The module uses different convolutional kernels to process street images of different scales. Thus, it obtains more comprehensively the different resolution features of various targets in complex street images. Finally, we use the double decoder to recover the details of the image in more detail to obtain the pixel-by-pixel classification results. The experimental results show that the network in this paper has higher segmentation accuracy on the Cityscapes dataset compared with other excellent networks such as DeepLabV3 and SegFormer. The mean intersection over union reaches 80.11% and an improvement of more than 3.51% compared to other networks. The method provides better understanding of images of complex street scenes.
关键词:semantic segmentation;target area enhancement;attention mechanism;feature pyramid grid;multi-dimensional features
摘要:To address the problems of difficult street object recognition and slow segmentation due to low visibility in foggy scence, a foggy cityscapes semantic segmentation algorithm incorporating self-supervised contrastive learning is proposed. The algorithm selects the lightweight network MobileNetV2 as the backbone network. Deep aggregation atrous spatial pyramid pooling module is designed and a deep separable convolution with dilation rate is used to replace the normal convolution to enrich feature diversity. Then, we increase the similarity of semantically similar pixels and maintain the distance between different semantic pixels by fusing the contrastive learning framework, so as to improve the model’s ability to represent and discriminate the detailed edges of small target objects. Finally, a new fusion loss function is proposed, and supervised learning and self-supervised learning are used to jointly guide the network training to learn deep feature representation. The experimental results show that the model can achieve MIoU of 74.35%, MPA of 83.59%, and PA of 95.85% on the Foggy Cityscapes dataset, which improves 3.82%, 3.99% and 1.02% respectively, compared with the semantic segmentation network DeepLabV3+ model. Meanwhile, the number of model parameters is 2.88M, which is nearly 55% less than the number of DeepLabV3+ model, optimizing the network computation consumption. The algorithm has good performance in foggy cityscapes semantic segmentation, reducing the number of model parameters while maintaining high segmentation accuracy and good robustness.
关键词:semantic segmentation;self-supervised learning;deep aggregation;contrastive learning;loss function