最新刊期
- 摘要：Relative radiometric normalization is a fundamental technique in multi-temporal remote sensing analysis， yet traditional methods often face limitations when dealing with nonlinear distortions， outlier interference， and heterogeneous land-cover conditions. To address these challenges， this paper proposes a diffusion-based probabilistic framework that models radiometric inconsistency as a combination of deterministic residuals and stochastic perturbations. Within this framework， the forward process simulates the multi-source characteristics of radiometric distortions through the superposition of structured and random noise， while the reverse process relies on a dual-objective variational optimization strategy to reconstruct radiometric consistency while preserving the detailed features of the original imagery， thereby achieving refined modeling of complex radiometric distortions at the mathematical level. Its core design is a spatial-spectral attention network， which integrates spatial and spectral attention modules to dynamically enhance key spectral band responses and local texture structures during feature extraction， enabling efficient capture of cross-band dependencies and multi-scale spatial context. To further improve the model's adaptability in complex scenes， a preprocessing strategy based on the structural similarity index （SSIM） is designed. By automatically selecting stable pseudo-invariant regions to guide model training， it effectively reduces interference from land-cover changes and enhances the representativeness of training samples and the stability of model convergence. Comprehensive experiments on multi-temporal Sentinel-2 datasets demonstrate that the proposed method consistently outperforms existing comparative approaches in quantitative evaluation metrics. It not only achieves higher radiometric correction accuracy and spectral fidelity but also exhibits significant advantages in vegetation index consistency and texture detail preservation. In summary， this diffusion-driven network provides a practically adaptable solution for large-scale relative radiometric normalization of remote sensing imagery and can be further extended to applications such as multi-sensor fusion and time series analysis in the future.关键词：relative radiometric normalization;diffusion model;attention mechanism;deep learning0|0|0更新时间：2026-04-15
- 摘要：Cholesteric liquid crystal elastomers （CLCEs） have attracted considerable attention in smart color-changing materials due to their unique helical structures and tunable optical responses. However， the thermal color-change range of CLCEs is often limited by the constraints of the polymer network， resulting in a relatively narrow color-change range or requiring elevated temperatures to achieve significant optical variation. In this work， a series of cholesteric liquid crystal films （CLCFs） with varying crosslinking densities were prepared by introducing different amounts of the chain extender EDDET. The influence of crosslinking density on both thermochromic and mechanochromic behaviors was systematically investigated. Increasing the EDDET content effectively reduces the crosslinking density， leading to a significantly broadened thermochromic range. Within 20~100 °C， a maximum red shift of 259 nm in the reflection wavelength was observed. Meanwhile， the thermal sensitivity was markedly enhanced， reaching a local maximum of 8.2 nm ℃-1 in the 60~80 ℃ range. Mechanical tests reveal that lower crosslinking density improves the elongation at break and enlarges the strain-induced blue-shift range. After pre-stretching treatment， the thermally induced wavelength shift of the CLCF can be extended to 340 nm， with an average thermal sensitivity of 4.31 nm ℃-1. These CLCFs， featuring a wide thermochromic range and high sensitivity， are promising candidates for smart sensing， information encryption， and wearable optical devices.关键词：cholesteric liquid crystal;polymer;thermochromism;optical sensing0|0|0更新时间：2026-04-15
- 摘要：To address the problems of local texture and edge-detail loss， limited receptive fields in the convolution branch， and insufficient use of cross-layer structural information in the graph branch for hyperspectral image classification， this paper proposes a CNN-GCN joint classification model with high-low frequency enhancement and cross-layer graph convolution aggregation. The model improves input representation through high-low frequency residual enhancement， extracts multi-scale spectral-spatial features using a multi-stage dynamic convolution encoder， enhances regional structural modeling by cross-layer weighted aggregation in the graph branch， and performs collaborative modeling through cross-branch attention fusion. Experiments on three public datasets， Indian Pines， Pavia University， and Salinas， achieve overall accuracies of 92.94%， 95.11%， and 97.50%， with corresponding Kappa coefficients of 91.94%， 93.50%， and 97.22%， respectively. The results show that the proposed method can effectively integrate local details， spatial context， and regional topological structure information， achieving competitive classification performance across different types of hyperspectral classification scenes.关键词：Hyperspectral image classification;spectral-spatial joint features;graph convolution network;attention fusion0|0|0更新时间：2026-04-15
- 摘要：ObjectiveMaterials science research is trying to develop deep learning-based computer vision methods， but currently limited experimental data is difficult to support the exploration of such big data-based methods. However， the limited experimental data cannot support such kind of big-data based research. This paper proposes an improved boundary balanced generative adversarial network （BEGAN） data augmentation model to overcome this issue.MethodFirstly， changing the normalization method in the generator network to spectral normalization reduces the requirement for training sample size compared to batch normalization； Secondly， adding residual modules to the generator/decoder of the model avoids overfitting and accelerates model training； Finally， a self-attention mechanism is added to enhance the model's ability to extract defect details， resulting in smoother and faster convergence of loss parameters during the training process.ResultThis paper conducted ablation experiments and comparative experiments using a publicly dataset of steel defects. Through two evaluation metrics and classification network accuracy， the experiments demonstrated that the improved model significantly outperforms four mainstream generative models. Compared to the BEGAN model's generative dataset， the image classification performance improved by 5.55%； the FID value decreased by 54.35%； the IS value has increased by 18.18%.ConclusionThe performance of generated specimens proved this improved method is sufficient as an image enhancement method to cope with small sample problems.关键词：steel surface defect;data augmentation;neural network;generative adversarial network0|0|0更新时间：2026-04-15
- 摘要：To address the challenges faced in citrus detection on fruit trees in natural environments， such as difficulty in maturity differentiation， severe occlusion from branches and leaves， high model complexity， and resource deployment limitations， this paper proposes a lightweight citrus maturity detection algorithm for fruit trees， YOLO-HiP， based on an improved YOLOv11. First， an improved HGNetv2-L network is used as the backbone， combined with a hierarchical feature extraction strategy， significantly enhancing the model’s capability to analyze complex scenes while effectively reducing computational complexity and resource consumption. Next， a lightweight hybrid attention module， C2PSA_iRMB， is designed. By integrating the C2PSA mechanism with the iRMB module， the computational cost is optimized， and the ability to process long-range information is enhanced， improving the module’s flexibility and computational efficiency. Finally， a C3k2_PConv module is constructed， further improving spatial feature extraction efficiency by reducing redundant computations and memory access. Experimental results show that YOLO-HiP achieves 94.3% mAP50， an improvement of 4.7% over the original model， with only 5.1M parameters （a 45.7% reduction）， a computational load of 13.9 GFLOPs （a 34.7% reduction）， and a frame rate of 227.4 FPS （a 25.1% increase）. This model significantly reduces model size while ensuring detection accuracy， providing an innovative and feasible solution for platforms with limited computational resources， such as citrus-picking robots and other embedded systems.关键词：citrus;maturity;YOLOv11;lightweight;object detection0|0|0更新时间：2026-04-15
- 摘要：To address the limitations of existing shadow removal methods， such as insufficient modeling of global contrast relationships， limited regional differentiation processing capabilities， and the tendency to produce shadow residuals and boundary artifacts， a shadow removal network based on regional contrast and boundary perception is proposed.Firstly， a multi-scale attention mechanism is designed to enhance the consistency of image structure by utilizing complementary spatial representations and key region enhancement. Secondly， a shadow perception module is constructed to apply differentiated feature processing strategies to shadow and non-shadow regions， improving the continuity of shadow boundaries and their transition areas. Then， a visual contrast attention mechanism is proposed to enhance the ability to define cross-regional contrast relationships at the global scale. Finally， a joint loss function is designed， with a dynamic weight adjustment strategy to guide the model to achieve a coordinated optimization of pixel accuracy， structural consistency， and color stability.The experimental results show that， compared with 8 representative methods on the ISTD dataset， the PSNR is improved by an average of 1.57 dB， and compared with 10 representative methods on the SRD dataset， the PSNR is improved by an average of 2.03 dB. Meanwhile， with 12.3M parameters， the proposed method achieves an effective balance between reconstruction quality and model complexity.This method significantly reduces pixel-level brightness errors and effectively suppresses shadow residuals and boundary artifacts， achieving a more natural and reliable shadow removal effect.关键词：shadow removal;Multi-scale attention;shadow aware;visual contrast attention;cross-regional comparison2|0|0更新时间：2026-04-15
- 摘要：Leakage of light is a common phenomenon when a LCD（Liquid Crystal Display） screen displays a black pattern， mainly caused by the combination of the screen structure and the characteristics of the backlight system. A common and simple yet effective way to improve this issue is to reduce the overall screen brightness， which can significantly decrease the visibility of the leakage. This paper systematically studies the variation law of the liquid crystal response time after activating the dynamic backlight control， and measures the liquid crystal response time data under black and white pattern with different PWM（Pulse With Modulation） duties output by the TCON（Timing Controller）. This is achieved through the DBC（Dynamic Backlight Control） function in the TCON， where the PWM duty output by the TCON is 40% of the input PWM when displaying a black pattern， and 100% of the input PWM when displaying a white pattern to the LED（Light Emitting Diode） driver. This method can improve the leakage of the black pattern， but in DC（Direct Current） dimming applications， combined with the mechanism of the LED driver identifying the PWM duty and then outputting the LED current， it will lead to an increase in the liquid crystal response time. The test result of the optical instrument is 51.9 ms， which seriously exceeds the design specification of Max 25 ms. To address this issue， this paper proposes setting the PWM duty for the white pattern output to 99.6% while maintaining the maximum brightness of the LCD screen unchanged. The measured liquid crystal response time can be shortened to 24.7ms， which not only solves the problem of black pattern leakage but also meets the liquid crystal response time specification.关键词：LCD;LED;dynamic backlight control;liquid crystal response time27|10|0更新时间：2026-04-03
-
Analysis of research status and development trends in oxide semiconductor thin-film transistors AI导读
摘要：Oxide semiconductor thin-film transistor （OS TFT） has become a core driving component of high-end display technology due to its advantages such as high mobility， excellent uniformity and low manufacturing cost， and is gradually expanding into integrated system fields such as sensors， memory， and neuromorphic computing. Based on the Web of Science and Incopat databases， this paper conducts a systematic analysis of the scientific research papers and patent data in the field of OS TFT from 2016 to 2025. Research has found that global R&D activities are highly concentrated in East Asia， with China and South Korea leading in the output of papers and patents. However， there is a clear academic-industry imbalance in this field： over 90% of the papers come from universities and research institutes， while 81.01% of the patents are held by enterprises， and the transformation of research into applications is insufficient. Meanwhile， the concentration of patents is relatively low， and technological barriers have yet to be formed. The key upstream material， sputtering targets， are still dominated by Japan （accounting for 54.55%）， posing a potential risk to the security of China's industrial chain. In the future， OS TFT is poised to move beyond conventional display driving， forging deep convergence with sensor， memory， and neuromorphic computing.关键词：oxide semiconductor thin-film transistors;display;sensor;memory;neuromorphic computing24|26|0更新时间：2026-04-03 - 摘要：To address the voltage gradient and phase distribution non-uniformity in traditional comb-electrode liquid crystal（LC） lens， this study designs a electrode structure to improve optical performance. By optimizing the electrode layout without increasing the number of driving voltages， uneven voltage distribution between adjacent electrodes is resolved. The new and conventional electrode LC lens samples were fabricated via photolithography. Their wavefront patterns and root-mean-square （RMS） errors were then analyzed through interferometry. Finally， the phase distribution and imaging performance of the LC lenses were analyzed. The results show that the new electrode structure eliminates voltage gradient unevenness. The wavefront pattern exhibits a regular circular distribution， with wavefront RMS reduced by approximately 19.4% compared to the conventional structure. Both phase distribution and imaging performance are improved. This study confirms the effectiveness of electrode structure optimization in enhancing LC lens performance， offering a technical approach for developing high-precision tunable-focus optical devices.关键词：liquid crystal lens;electrode structure;phase distribution;parabola12|18|0更新时间：2026-04-03
- 摘要：Infrared and visible image fusion aims to retain the complementary features of different modalities to achieve robust perception of complex scenes， and it plays a crucial role in numerous fields such as security monitoring， military reconnaissance， and autonomous driving. However， existing image fusion algorithms focus on enhancing the visual effects of images， leading to the ineffective preservation of key semantic information during the fusion process， which in turn affects the application performance of fused images in high-level visual tasks. Although existing methods attempt to cascade the fusion task with high-level visual tasks （segmentation， detection， etc.）， this sequential connection has limited enhancement on semantic information. To balance visual effects and downstream tasks， this paper proposes a semantic-driven infrared and visible image fusion network， SDFusion. First， a shared feature encoder is adopted to perform multi-level cross-modal feature extraction for infrared and visible images， fully mining the useful information in both modal images. Then， through the parallel collaborative optimization of the image fusion decoder and the semantic segmentation decoder， while hierarchically injecting encoded features into decoded features to enhance feature representation， the joint modeling of fusion features and semantic features is achieved. Experimental results on public datasets show that compared with traditional methods， this method significantly improves seven objective evaluation indicators. Specifically， EN is improved by 3.7%， SD by 7.3%， MI by 45.3%， VIFF by 18.5%， and
- 摘要：Aiming at the problem that the stitching of borehole wall images in geological exploration is easily affected by multiple factors such as imaging illumination conditions， probe disturbance， and information missing of depth or azimuth， a robust stitching method for borehole wall images under multi-interference imaging conditions is proposed. Firstly， an image mapping method with conditional constraints is adopted to expand a specific sampling region of the borehole wall image into a narrowband image that can be effectively used for stitching under the interference of the imaging light source. Then， the intrinsic characteristics of the images in borehole video are utilized， and an ROI-based registration strategy is adopted to realize the accurate registration of the unfolded narrowband borehole images under the conditions of probe disturbance and information missing of depth and azimuth， and then a precise registration of the unfolded narrowband image of the borehole under the conditions of probe disturbance， missing depth or azimuthal angle. At the same time， a grid-based matching strategy is applied to improve the registration speed. Finally， a row-by-row dynamically weighted image fusion method is adopted to eliminate the stitching traces according to the registration results， which further optimizes the quality of the stitched images. Through the experiments on the simulated images under different geological conditions， the performance indexes of the proposed method with average value of SSIM of 71.39%， average value of PSNR of 26.74， and average value of registration accuracy of 92.38% are better than those of the comparison method. and through the experimental results on the real images of different borehole videos， the objective evaluation indexes of the stitching results obtained by the proposed method， such as EN， SF， AG， MI， and Q_MI， etc.， are all improved.关键词：image stitching;image registration;image fusion;borehole wall image163|288|0更新时间：2025-02-25
- 摘要：Manganese （Mn2+） ion-doped all-inorganicCsPbCl3 perovskite quantum dots have shown great potential for application in Mini/Micro-LED display devices due to their unique photoluminescence properties and nano-scale particle size. However， the current photoluminescence quantum yields of CsPbCl3：Mn2+ perovskite quantum dots is low and cannot meet the needs of practical applications. Herein， the potassium （K+） and Mn2+ ions co-doped CsPbCl3 perovskite quantum dots are synthesized via the hot injection method. The prepared CsPbCl3：（K+， Mn2+） perovskite quantum dots present a dual-color emission， which can be assigned to the perovskite excitons emission and Mn2+ ions emission. The photoluminescence quantum yield of CsPbCl3：（K+， Mn2+） perovskite quantum dots is improved from 16.5% of original to 66.2% through the optimization of K+ ions doping concentration. Meanwhile， the regulation mechanism of K+ ions doping on the photoluminescence properties of CsPbCl3：Mn2+ perovskite quantum dots are investigated. It is shown that the incorporation of K+ ions can effectively inhibit the formation of intrinsic defect states and Mn-Mn dimers or Mn-related defects in CsPbCl3：（K+， Mn2+） perovskite quantum dots， thus enhancing the radiative recombination luminescence of carriers in perovskite quantum dots. Benefiting from the strategy of hetero-ions doping， these doped perovskite quantum dots are expected to be applied in Mini/Micro-LED display fields.关键词：CsPbCl3： Mn2+ perovskite quantum dots;K+ ions doping;dual-color emission;photoluminescence quantum yields192|231|0更新时间：2024-09-04
- 摘要：Polymer dispersed liquid crystal （PDLC） is a thin film formed by liquid crystal （LC） droplets embedded in a continuous polymer matrix in the form of micron-sized droplets. The use of click chemistry method to prepare PDLC can avoid the problems of dye decomposition caused by traditional photopolymerization-induced phase separation. In this study， four polyethylene glycol diacrylates （PEGDA） of different alkyl chain lengths were used as raw materials to synthesize PDLC films through thiol-ene click reaction and thiol-isocyanate coupling reaction， to study the effects of different alkyl chain length olefin monomers on the electro-optical properties of PDLC films. The results showed that as the chain length of olefin alkyl chain in the system increased， the threshold voltage （Vth） ， saturation voltage （Vsat） and contrast ratio （CR） decreased， while off-state transmittance （Toff） and on-state transmittance （Ton） increased. By selecting olefin with appropriate alkyl chain length， PDLC films with good electro-optical properties can be prepared.关键词：polymer dispersed liquid crystal;click reaction;monomer alkyl chain length;electro-optical property269|293|0更新时间：2024-09-04
- 摘要：Virtual reality technology has huge application potential in many fields， and panoramic cameras are one of the important tools to realize virtual reality.Aiming at the problem that occlusions such as brackets occlude the bottom image of a 720° panoramic camera， this paper proposes a method for eliminating the occlusion of the bottom bracket based on step-by-step shooting. Eliminate bottom occlusion by shooting the side and top once and shooting twice on the bottom.On this basis， optimized for the inconsistency of image brightness on the side， top， and bottom.The global uniformity algorithm will cause images in different regions to influence each other， causing deviations in the uniformity results.Therefore， this paper proposes a Wallis uniform color algorithm based on region growing method image segmentation.The experimental results show that the method in this paper can eliminate the bottom image occlusion； using the algorithm in this paper to even color， the image color difference is reduced by 59.3% compared with the global algorithm.关键词：panoramic camera;panoramic image stitching;dodging algorithm130|246|0更新时间：2024-09-04
- 摘要：Aiming at the problem of large scene， large field of view image applications and the slow speed of traditional image stitching algorithms， a fast image stitching algorithm based on offline calibration is adopted. In practical applications， the positions of multiple cameras are fixed， and offline calibration is used to calculate the homography matrix of image--stitching， which saves a lot of feature extraction and registration time. On the basis of fast image alignment， this paper proposes a fast image fusion algorithm based on the Laplacian pyramid of YUV color space， and it has better fusion effect and faster speed than the traditional Laplacian pyramid fusion algorithm. Experimental results show that The image fusion speed of the article algorithm is three times that of traditional Laplacian image fusion. Meet the real-time image stitching speed requirements in application scenarios， with good image stitching effects.关键词：image stitching;Laplacian pyramid;feature extraction;fast fusion274|335|0更新时间：2022-11-07
- 摘要：In order to solve the problem of dark color and incomplete defogging after image defogging， an image defogging algorithm based on multi-scale concat convolutional neural network is proposed in this paper. Taking the foggy image as the input， the shallow layer information of the image is extracted from the single scale convolution layer through the preprocessing module， and then the multi-scale mapping module is designed to realize the depth feature learning and the fusion of the deep and shallow layer features. The deconvolution module is used to restore the image size， and the coarse transmittance map corresponding to the foggy image is obtained through the convolution operation. Finally， the haze free image is restored according to the atmospheric scattering model. The experimental results show that the proposed method is superior to other algorithms in both synthetic and natural foggy images， and the peak signal-to-noise ratio （PSNR） and structure similarity （SSIM） can reach 29.238 and 0.9502 respectively. The proposed algorithm can effectively avoid the dark color and distortion of the image， improve the image defogging performance and show good visual effect.关键词：image dehazing;convolutional neural network;multi-scale concat;image restoration;atmosphere scattering model220|196|0更新时间：2022-11-07
- 摘要：Nowadays， micro-LED is considered as a crucial and one of the fastest growing display technologies in the world as it finds its applications in variety of products from visible light communication applications to large flat panel displays， virtual reality and wearable displays， televisions and light sources for the optogenetics and neural interface. Though the prospects are bright， micro-LEDs still face some technological problems which needs to be addressed in order to get high volume commercialization， which include improving efficiency of LEDs with longer wavelengths， improving efficiency at low current densities， full color schemes， mass transfer， defects and yield management， repair technology and cost control. In this review we have highlighted the different challenges and their optimum solutions for micro-LEDs.关键词：Micro-LEDs;Display technologies;efficiency of LEDs;mass transfer;III-Nitride566|1077|0更新时间：2022-11-07
- 摘要：In view of the problem that the quality of high dynamic range image appearance needs to be improved， In this paper，a tone mapping algorithm based on collaborative filtering is proposed in the chromaticity brightness color space. First， the input high dynamic range image uses the chromaticity brightness color space to extract brightness information and chromaticity information， and then decomposes and reconstructs the chromaticity information and brightness information respectively. Using bilateral filtering technology to decompose the brightness information to obtain the brightness basic layer and brightness detail layer， according to the consistency of the large-scale edge of brightness information and chromaticity information， an algorithm of chromaticity brightness collaborative filtering is constructed to break down the chromaticity information， and obtain the chromaticity base layer and chromaticity texture layer. Finally， the brightness information and chromaticity information are reconstructed， and then converted to the RGB color space to output the final tone mapping results. The results of objective index calculation show that the results of this paper are improved by 25.24%，18.89%，45.89% respectively in image quality score， structural fidelity and image naturality， so that boundary details and color information can be better maintained.关键词：high dynamic rang image;tone mapping;chromaticity brightness color space;bilateral filtering;collaborative filtering331|676|0更新时间：2021-09-27
- 摘要：In order to assist R & D investment and science and technology projects management， the evolution trend of some main display techniques was evaluated.Based on the S-curve evolution trend， new display technology systems' development stage and direction were judged. And the relationship between R & D investment and market life was analyzed by using the patent quantity in different years.It is showed that liquid crystal and organic light-emitting diode display systems are in the mature and growing period， respectively， E-paper and laser display are in the transitional period， quantum dots and micro light-emitting diodes are in the infancy. A large number of patents indicates a large R & D investment and long market life of a technology. The lower the maturity of the technology system， the higher the invention level of R & D achievements. Therefore， the investment in basic research should focus on infancy， and the investment in R & D of common key technologies should focus on the transition period and the early growth period. The content and leading investors of R & D were put forward.关键词：evolution trends;S-curve analysis;display system;R&D investment1446|457|3更新时间：2021-08-05
- 摘要：Simultaneous location and mapping （SLAM） is a key technology for autonomous driving vehicles and robots to realize autonomous movement. The lidar currently widely used in SLAM technology present a series of issues， including high cost， low spatial resolution of laser point clouds， and difficulty in obtaining accurate semantic information. In contrast， cameras can effectively avoid the above problems， but more complex algorithms are required in depth prediction and mapping. In recent years， with the increase in computing power， the continuous enrichment of data sets， as well as the introduction of new machine vision algorithms， the accuracy and efficiency of vision depth prediction and mapping algorithms have been greatly improved. This article summarizes the existing methods of visual depth prediction and point cloud mapping and classifies the methodology based on visual data collection approach and algorithm design， then analyzes the application scenarios and prospect of visual depth estimation and point cloud mapping.关键词：monocular depth detection;binocular depth detection;deep learning;point cloud mapping82|102|0更新时间：2021-04-09
- Address：No.3888 Dong Nanhu Road, Changchun, Jilin, China 130033 Postal code：130033
- Tel：0431-86176059 Email：yjxs@ciomp.ac.cn
- Technical support is provided by Beijing Founder electronics co., LTD 吉ICP备11002662号-17
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
0