To accurately capture small defect information in insulators,improve the reconstruction quality of terahertz images,and achieve precise insulator defect detection,a method for insulator defect detection based on terahertz pulse time-domain feature reconstruction algorithm was studied.This method combined improved Hilbert Huang transform and ensemble empirical mode decomposition to perform deep analysis of terahertz time-domain signals of insulators and accurately extract defect features.Multi-dimensional features were effectively fused through principal component analysis,and they were mapped to an orthogonal basis feature space using coordinate transformation techniques,a two-dimensional terahertz image was reconstructed that contain rich feature information of insulator defects.Furthermore,the reconstructed image was input into the YOLOv3 based insulator defect detection model,where the Darknet-53 feature extraction network was used to capture defect features,achieving intelligent defect detection.The experimental results showed that the proposed method can effectively improve the quality of generated terahertz images of insulators and these images were rich in defect feature.The proposed method achieved a recognition rate consistently above 0.95 and could reach up to 0.98,demonstrating high detection accuracy.
As one of the important features of biometrics,fingerprints are widely used in criminal investigation.However,the traditional fingerprint acquisition technology is often limited to two-dimensional contact measurement,which is easy to be affected by finger humidity,force deformation and other factors,resulting in poor recognition effect.Therefore,for fingerprint measurement,this paper designs a small field of view imaging system based on structured light was.By adding lenses and other ways,the imaging field of structured light is reduced,and high-precision three-dimensional fingerprint information is obtained by non-contact measurement.At the same time,due to the influence of equipment accuracy,environmental noise and other factors,the acquired fingerprint point cloud usually contains noise.This paper analyzes the source of the acquired fingerprint point cloud noise,classifies the noise,proposes a multi-scale noise denoising method based on statistical analysis and Gaussian filter,and verifies its effectiveness through experiments.
Aiming at the problems of color distortion,low contrast and detail blur caused by light absorption and scattering in water,an underwater image enhancement algorithm based on red channel prior was proposed.Combining the traditional red channel prior theory with the generation adversarial network,a red channel prior module was designed to pay attention to the attenuation characteristics of the red channel in underwater images and conduct joint training with the encoder of the generator,so that the encoder could use the information provided by the red channel prior module to extract useful features better,which was conducive to solving the color bias problem of underwater images.The discriminator adopted global-local double discriminator,and constructed several loss functions to make the generated image consistent with the reference image in structure,content and color.Compared with other 10 algorithms,the UIEB,EUVP and LSUI public data sets obtained optimal or sub-optimal results.The enhanced underwater image peak signal-to-noise ratio(PSNR),structural similarity(SSIM),underwater image quality metric(UIQM),underwater color image quality assessment(UCIQE)and information entropy(IE)on the UIEB dataset were increased by 1.78 dB,0.073,0.016,0.004 and 0.02,respectively.Mean square error(MSE)decreased by 48.63.The algorithm proposed in this paper has achieved remarkable results in improving the quality of underwater images,showing obvious advantages in both subjective visual perception and objective data.
Phase-only hologram has been widely noticed in the field of holographic display due to the characteristics of having no conjugate scattered light and almost no optical loss in the imaging process.Computer-generated holograms based on deep learning have great potential for generating high-efficiency and high-quality display holograms,and have an important position in the field of phase-only hologram generation.This paper introduces computational holography technology,traditional phase-only hologram generation algorithms and model-driven deep learning fundamentals,overviews deep learning model-driven computational holography solutions based on deep learning proposed in recent years,compares optimization methods based on the network structure and loss function,and looks forward to the development and challenges of deep learning technology in the field of computational holography.
Aiming at the quality degradation problems such as low brightness and contrast,non-uniform illumination of images captured in real environments,a non-uniform low illumination image enhancement algorithm based on Retinex theory was proposed.Firstly,the original image was converted from RGB(red,green,blue)space to HSV(hue,saturation,value)space by the algorithm,and the V-component was extracted for enhancement processing.The Retinex decomposition was implemented by the algorithm through combined window filtering,with side window filtering and full window filtering being used for the edge and texture of the image respectively to obtain a locally smooth and structure-preserving illuminance component.For the illuminance component,a luminance transformation method based on the quadratic curve is used for indirect adjustment to effectively enhance the low illuminance and correct the non-uniform illuminance.For the reflection component,hybrid filtering was used to suppress noise and sharpen edge details.Finally,the enhanced illumination component and reflectance component were recombined to obtain the enhanced V-component,and then converted back to RGB space to obtain the final enhanced image.The experimentalresults show that the NIQE(natural image qualityevaluator),NIQMC(no-reference image quality metric),CEIQ(con-tent-enhanced image quality)and Entropy metrics for this method are 2.747,5.380,3.432 and 7.476,respectively,which are superior to most existing image enhancement algorithms.The proposed algorithm not only enhances the brightness and contrast,but also effectively corrects the non-uniformity of image illumination,making the visual effect clearer and texture details richer.
The optical properties of HfO2 thin films are greatly influenced by the preparation process factors.Using hafnium metal as the raw material and oxygen as the reaction gas,the monolayer HfO2 thin films were deposited on quartz glass substrates by electron-beam thermal evaporation technology.The effects of evaporation current,working pressure and deposition temperature on its optical properties and laser-induced damage threshold were studied through orthogonal experiments.The spectral transmittance of monolayer HfO2 thin films were measured by a spectrophotometer,its refractive index and extinction coefficient were measured by an ellipsometry,and its laser-induced damage threshold was measured by laser-induced damage test system.The research finds out that the evaporation beam is the most important factor affecting the optical properties and laser-induced damage threshold of monolayer HfO2 thin films,the working vacuity is the secondary factor,and the influence of deposition temperature is not significant.The optimal process parameters affecting the optical properties of HfO2 thin films were obtained through the process optimization,The evaporation beam is 170 mA,the working pressure is 1.8×10-2 Pa,and the deposition temperature is 160℃.The prepared HfO2 thin films at this process parameters have a refractive index of 1.889 6 at 1 064 nm,an extinction coefficient of 3.07×10-5 at 1 064 nm,and a laser-induced damage threshold of 23.1 J·cm-2 at a single pulse energy of 1 064 nm.
In order to achieve accurate detection of the outdoor infrared ranging system in a limited confined space,the volume of the optical system needed to be compressed on the basis of not affecting the detection results,and a set of foldback optical paths with a total optical length of 4 m and a maximum mirror of 12.8 cm was designed using a two-dimensional biplane reflector system.The foldback optical path was divided into 2×(n+1)sections with n planar mirrors,and the number of mirrors and the optical diameter were optimized using the Global Search algorithm,so that each detection ray could irradiate the detected object and be successfully folded back to the detector.After meeting the detection requirements,the theoretical verification analysis was carried out by using the signal wave principle to avoid the detection errors caused by the optical range loss.Finally,by analyzing the simulation results of stray light of the system in a confined space,the stray light suppression of the system was carried out,and the experiments were compared with the detection results of the original path to verify the correctness and accuracy of the design.
As a vital research direction of computer vision in the fields of national defense and civil monitoring,infrared dim and small target detection acts as a key technical support for tasks such as long-range early warning,maritime surveillance and environmental perception of unmanned equipment,which is of great significance for improving the environmental adaptability and task response accuracy of perception systems.Aiming at the problems of low signal-to-noise ratio,background clutter interference and inconspicuous target features in infrared dim and small target detection under complex scenarios,an improved infrared dim and small target detection model with edge-aware gating and three-path adaptive fusion was proposed.Firstly,a U-shaped encoder-decoder architecture was adopted to realize multi-scale semantic fusion,with Swin-Transformer as the backbone to uniformly model local details and long-range dependencies.Then,to effectively suppress background noise and enhance target edge responses,a collaborative module integrating the recursive enhanced edge-semantic gated contextual layer(REES-GCL)and atrous spatial pyramid pooling(ASPP)was improved and designed,which ccould adaptively suppress complex background clutter and accurately extract target edge information.Finally,addressing the issues of significant semantic gap and lack of targeted weight allocation in traditional feature fusion,a gated dual-stage attention fusion(GDSAF)module was improved and designed to achieve the precise collaboration of three-path features.Experimental results show that the proposed model achieves a significant improvement in robustness under complex backgrounds and low-contrast scenarios,with the detection precision and recall rate increased by 7.3%and 3.1%respectively,compared with the state-of-the-art comparison model,which verifies the effectiveness of the collaborative mechanism of REES-GCL and ASPP as well as the dual-stage attention fusion module.
To address the issues of high computational resource demands and low accuracy in existing defect detection algorithms for LED bulb appearance defect detection,an RSME-YOLO algorithm wasd designed.First,a reconstructed lightweight RD-HGNet backbone was proposed to reduce computational redundancy while enhancing gradient stability.Second,a Slim-Neck module was adopted to minimize redundant computations while preserving cross-channel interactions,efficiently maintaining multi-scale feature fusion.Then,a MSA-Detect detection head incorporating multi-head self-attention mechanism was designed to enhance feature interaction capability and model expressiveness.Finally,by replacing the loss function with the decoupled EIoU loss that separately handles length and width,the negative impact of unbalanced optimization on the model was resolved.Compared to the baseline YOLOv8n algorithm,RSME-YOLO reduced the parameter count by 33.2%,lowered GFLOPs by 40.7%,and had a model size of only 4.2 M.It achieved mAP50 scores of 87.1%and 86.8%on the validation and test sets,respectively.With higher detection accuracy and lightweight design,it is more suitable for intelligent upgrades in LED defect detection for small and medium-sized enterprises.
In random lasers,fiber Bragg gratings(FBGs)and cascaded multi-FBG configurations are extensively employed as wavelength-selective components.However,the modulation depth of FBGs exerts a non-negligible influence on the output performance of random lasers.In this study,two cascaded FBGs were taken as the research subject,and the effect of discrepancies in FBG modulation depth on the output spectrum of random lasers based on Rayleigh scattering feedback was comparatively analyzed.When the modulation depths of the two FBGs were 3 dB and 6.3 dB,respectively,only the FBG with a modulation depth of 6.3 dB enabled random lasing at its central wavelength due to gain competition.When the modulation depths were 5.5 dB and 6.3 dB,respectively,stable dual-wavelength random lasing was achieved;within a 1-hour period,the wavelength drifts were 0.060 nm and 0.044 nm,while the peak intensity fluctuations were 1.439 dB and 1.486 dB,respectively.The findings of this research can provide valuable insights for the design of wavelength-selective components in random lasers and the investigation of random laser performance.
Spatial heterodyne spectroscopy is a new Fourier transform spectroscopy technique,which is widely used in the field of atmospheric remote sensing because of its high sensitivity and high resolution.The interferogram data obtained by the space heterodyne spectrometer is often accompanied by noise interference,which makes the measurement results appear to be wrong.Based on this,we proposed a spatial heterodyne interferogram noise reduction method based on 2D frequency domain filtering.By analyzing the characteristics of measured 2D spectrum diagram of potassium lamp and the simulated ideal 2D spectrum diagram,the threshold data was introduced to construct a specific filter,so as to separate the effective signal and the noise signal,and compare with the average spectral data of the interferogram before and after noise reduction.The application effect of the treatment method was evaluated.The proposed method in this paper was applied to the noise reduction processing of the interferograms from 3 light sources:potassium lamp(quasi-monochromatic light),xenon lamp,and water vapor(continuous light).The results show that after noise reduction of the potassium lamp interferogram,the characteristic peaks in its transformed spectrum are highlighted and the surrounding noise is significantly reduced.After noise reduction of the xenon lamp and water vapor interferograms,the dark spots in the interferograms are suppressed and the non-uniformity is improved.By comparing the spectra of 3 rows before and after correction with the average spectrum,the root mean square errors decreased by 80.61%,76.21%,78.45%and 78.59%,80.04%,87.72%,respectively;the peak signal-to-noise ratios increased by 44.74%,38.58%,41.03%and 39.10%,41.98%,54.59%,respectively.Compared with the results of the rotation filtering algorithm and the wavelet algorithm,the root mean square error of the xenon lamp is optimized by 68.08%and 54.36%,and the peak signal-to-noise ratio is optimized by 27.84%and 16.32%.The corresponding values for water vapor are 78.79%,80.40%and 38.83%,41.56%.The noise reduction effect is obvious,indicating that the algorithm proposed is feasible for the application of noise reduction in space heterodyne spectral data.
Few-shot semantic segmentation aims to perform pixel-level classification tasks under conditions of limited annotations.To further enhance the generalization ability of few-shot semantic segmentation based on prototype networks for unseen classes,we proposed a method based on self-correlation reinforcement and prototype supervision,addressing the issues of appearance discrepancies between support samples and query images and poor prototype quality.Firstly,we designed a self-correlation reinforcement module that leveraged the self-correlation of pixels within the query image to transfer initial auxiliary priors to the query data,generating high-level class prototypes that provided reinforced prior information with high discriminative power.Secondly,we introduced multiple progressive supervision losses,using the prototype's ability to recover the support mask as a supervision indicator for prototype quality.This allowed for self-regularized updates of the prototypes and self-matching updates of the auxiliary priors,effectively enhancing the prototype's ability to generalize from the support information and encouraging the auxiliary priors to retain more query-relevant details.The proposed method was validated on the few-shot benchmark dataset PASCAL-5i.Experimental results show that,under the 1-shot setting,the method achieves an mIoU(mean intersection over union)of 64.4%and an FB-IoU(foreground-background intersection over union)of 73.5%,demonstrating its effectiveness and advanced nature.
To address the issue of difficulty in measuring the pulse parameters of current 100 GHz bandwidth optoelectronic devices,an electro-optic sampling system based on a cage structure was proposed.The working principle of the electro-optic sampling method was introduced,the technical difficulties of the full-space electro-optic sampling method were analyzed,and the optical path structure in a targeted manner was optimized.A cage structure was introduced to integrate the laser beam splitting node and the illumination optical path,and the fiber-space light hybrid transmission method was adopted to improve the stability of the system,reducing the beam drift from 12 μm to 2 μm,reducing the system volume by 85%,and increasing the assembly and adjustment efficiency by 70%.Finally,the system was used to test the pulse width parameters of a 100 GHz optoelectronic detector.The results show that the pulse parameters of 100 GHz optoelectronic devices can be tested,and the repeatability is good.
In complex environments such as extreme lighting conditions,high-speed motion scenes,and limited computational resources,the detection accuracy of human pose estimation using frame cameras is prone to degradation due to overexposure and motion blur.The method of human pose estimation based on event point cloud by taking advantage of the high dynamic range,high temporal resolution of event camera was explored.Efficient event preprocessing was achieved by designing a representation of the event stream to the point cloud,combined with a fixed-time window sampling strategy.On this basis,the point cloud residual multilayer perceptron and self-attention were further fused to construct a multi-level feature extraction network structure to achieve the mapping of human joint point coordinates from 3D event space to 2D image plane.Experiments on the DHP19 event dataset showed that our method has significant results in the task of human pose estimation based on event data,with a low 2D mean per joint position error(MPJPE)of 5.91 pixel and a 3D MPJPE of 67.48 mm.
In order to meet the demanding external dimensions,working environment,and high image quality requirements of long range oblique photography,the optical system adopted a combination of Mersenne-type telescope and lens compensator operating at 0.486 μm～0.656 μm(EO channel)and 3.7 μm～4.8 μm(IR channel),respectively.A tilted beam-splitting plate between the primary and secondary mirrors was adopted to achieve dual band simultaneous imaging and compression of external dimensions.By utilizing the afocal char-acteristics of the Mersenne-type telescope,a secondary mirror was selected as the image motion compensation mechanism to achieve functions such as aberration correction,back-scanning compensation,and image stabilization.The transfer functions of each field of view in the system are all higher than 0.2 at the Nyquist frequency,and the distortions are all within±0.5%.The main system and lens compensator can be separately tested and adjusted,and then assembled to reduce the sensitivity of optical fabrication and adjustment.
To enhance the automation and positioning accuracy of the ocular axis measurement system based on low-coherence interferometry principle during the image acquisition stage,a lightweight image segmentation model for pupil center recognition was proposed.This model integrated an attention mechanism inspired by the Mamba state space modeling to enhance the feature representation of small target regions.It introduced frequency domain fusion and dual residual connections to improve the edge localization ability in skip connections;and it employed depthwise separable convolution to reduce model complexity and improve the performance of embedded deployment.Experiments were conducted using images collected by a Hikvision camera to build a dataset.The results show that the model achieves a 0.89%improvement in mIoU(mean intersection over union)compared to the SE(squeeze-and-excitation)module and a 1.05%improvement compared to CBAM(convolutional block attention module).It also improves the Recall index by 0.49%compared to TransUNet.The overall recognition accuracy reaches 99.86%,maintaining good stability in complex environments such as strong light interference and eyelash occlusion.Further,by combining Canny edge detection and least squares ellipse fitting algorithm,high-precision pupil center localization is achieved.This method provides a robust and accurate image processing solution for the ocular axis measurement system,significantly enhancing its automation and optical alignment capabilities.
Aiming at the problems of gas pressure fluctuation and background noise in the insulation state detection of gas insulated switchgear(GIS)equipment,a high sensitivity detection method for the insulation state of eco-friendly gases GIS equipment based on photoacoustic spectroscopy technology was proposed.The method used photoacoustic spectroscopy technology to suppress background noise and accurately detect the components of eco-friendly gases.Combined with the maximum correlation minimum redundancy criterion,the insulation state characteristics were optimized to improve the detection pertinence.High sensitivity detection was achieved by building a model with support vector machine.The detection process included photoacoustic signal detection,noise suppression,component concentration determination,feature selection and classification.Experiment results on C4/CO2 gas mixtures demonstrated a fitting accuracy of 0.978 for C4 compoment,and the dynamic response entropy index consistently remaining below 0.1,indicating that the proposed method has excellent detection sensitivity for monitoring the insulation state of eco-friendly gas GIS equipment,capable of accurately capturing the micro molecular movement changes.This provides a reliable basis for the early diagnosis of equipment insulation deterioration.
A scanning coherent lidar system based on rectangular grids was used to build a speed model of rotating targets.The theory of velocity features was studied,and its velocity spectrum matched the classic time-frequency spectrum.Velocity spectrum of precessing non-spinning cylindrical targets were calculated and simulated under different precession cone angles.Experiment results showed that simulation and experiment matched well.The proposed velocity feature spectrum enriches the speed model for rotating targets in scanning coherent lidar.It offers a new way to analyze and extract micro-motion parameters of rotating and precessing targets.
In order to satisfy the demand of biology,scientific research,industrialization and continuous wave frequency doubling ultraviolet laser,a research on blue diode laser dual end pumped Pr:YLF generating fundamental laser is designed.With the novel method of LD collimating and L-Shaped resonator of thermal lens compensation,a TEM00 red laser radiation at 639 nm with an output power as high as 21.82 W was obtained.To the best of our knowledge,this is the first report on a Pr:YLF laser operating at 639 nm with output power above 20 W.The result of this research satisfies the demand of high output power and the perfect beam quality at the same time.
