Pub Date : 2025-12-27DOI: 10.1016/j.optlaseng.2025.109566
Noritomo Okamoto , Shota Osada , Fan Wang , Tomoyoshi Ito , Tomoyoshi Shimobaba
Holographic displays require a significant amount of time to generate holograms. One of the methods used to solve this problem is the fast computation of Fresnel holograms using separable convolution. In this study, we propose a separable Fraunhofer hologram that can compute holograms even faster. In addition, we derive recurrence formulas suited to the proposed methods. We compare the computational speed and image quality of the proposed methods with those of the conventional method. Although a dedicated computer using a field-programmable gate array (FPGA) has been developed, the conventional algorithm has a problem of increasing the complexity of computational circuitry. In this study, we demonstrate that the proposed method can simplify a computational circuit.
{"title":"Accelerated hologram generation using separable Fraunhofer diffraction","authors":"Noritomo Okamoto , Shota Osada , Fan Wang , Tomoyoshi Ito , Tomoyoshi Shimobaba","doi":"10.1016/j.optlaseng.2025.109566","DOIUrl":"10.1016/j.optlaseng.2025.109566","url":null,"abstract":"<div><div>Holographic displays require a significant amount of time to generate holograms. One of the methods used to solve this problem is the fast computation of Fresnel holograms using separable convolution. In this study, we propose a separable Fraunhofer hologram that can compute holograms even faster. In addition, we derive recurrence formulas suited to the proposed methods. We compare the computational speed and image quality of the proposed methods with those of the conventional method. Although a dedicated computer using a field-programmable gate array (FPGA) has been developed, the conventional algorithm has a problem of increasing the complexity of computational circuitry. In this study, we demonstrate that the proposed method can simplify a computational circuit.</div></div>","PeriodicalId":49719,"journal":{"name":"Optics and Lasers in Engineering","volume":"199 ","pages":"Article 109566"},"PeriodicalIF":3.7,"publicationDate":"2025-12-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842191","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-27DOI: 10.1016/j.optlaseng.2025.109579
Fang Wang , Yuchang Wen , Guoqing Shangguan , Shuangshuang Han , Xinyi Zhao , Yanzhong Yuan , Hualei Shen , Yufang Liu
We design a fiber-optic stress system based on strong optical feedback from a Fabry–Perot laser diode (FP-LD). It monitors stress changes induced by tilting of high-voltage transmission towers. The system comprises an external resonant cavity formed between the rear facet of the FP-LD and the end face of a single-mode fiber (SMF). The effective optical range length of the resonant cavity varies, which leads to a shift in the beat-frequency signal (BFS) of the multiple longitudinal modes. In the 0–80 N applied stress loading experiments of the sensing fiber, the stress variation is linearly related to the frequency shift of the BFS. The system sensitivity increased with increasing sensing resonator length. Specifically, for resonator lengths of 3.0 m, 5.1 m and 7.2 m, the measured stress sensitivities were −6.70 kHz·N⁻¹, −9.95 kHz·N⁻¹ and −13.67 kHz·N⁻¹, respectively. Applying a multilayer perceptron (MLP) neural network reduced the error between the stress values predicted from the frequency-shift monitoring software and the true stress measurements. This method improves the detection accuracy of the system and its MLP model has an accuracy of 96.88% in the test set.
{"title":"FP-LD strong feedback fiber-optic stress system for high-voltage transmission towers","authors":"Fang Wang , Yuchang Wen , Guoqing Shangguan , Shuangshuang Han , Xinyi Zhao , Yanzhong Yuan , Hualei Shen , Yufang Liu","doi":"10.1016/j.optlaseng.2025.109579","DOIUrl":"10.1016/j.optlaseng.2025.109579","url":null,"abstract":"<div><div>We design a fiber-optic stress system based on strong optical feedback from a Fabry–Perot laser diode (FP-LD). It monitors stress changes induced by tilting of high-voltage transmission towers. The system comprises an external resonant cavity formed between the rear facet of the FP-LD and the end face of a single-mode fiber (SMF). The effective optical range length of the resonant cavity varies, which leads to a shift in the beat-frequency signal (BFS) of the multiple longitudinal modes. In the 0–80 N applied stress loading experiments of the sensing fiber, the stress variation is linearly related to the frequency shift of the BFS. The system sensitivity increased with increasing sensing resonator length. Specifically, for resonator lengths of 3.0 m, 5.1 m and 7.2 m, the measured stress sensitivities were −6.70 kHz·N⁻¹, −9.95 kHz·N⁻¹ and −13.67 kHz·N⁻¹, respectively. Applying a multilayer perceptron (MLP) neural network reduced the error between the stress values predicted from the frequency-shift monitoring software and the true stress measurements. This method improves the detection accuracy of the system and its MLP model has an accuracy of 96.88% in the test set.</div></div>","PeriodicalId":49719,"journal":{"name":"Optics and Lasers in Engineering","volume":"199 ","pages":"Article 109579"},"PeriodicalIF":3.7,"publicationDate":"2025-12-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842190","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-24DOI: 10.1016/j.optlaseng.2025.109543
Xiangfeng Xie , Ping Xu , Ji Xu , Wenjie Zhang , Yian Liu , Haifeng Zheng
Hyperspectral images (HSIs) captures more comprehensive spectral information than traditional RGB imaging, offering significant potential across various applications. Conventional methods, such as point, line, and wavelength scanning, are capable of generating hyperspectral images, but are often time-consuming and costly. Snapshot imaging systems, like the Coded Aperture Snapshot Spectral Imaging (CASSI), snap three-dimensional hyperspectral data into two-dimensional measuremen, enabling faster acquisition, reduced costs, and enhanced miniaturization. Despite these advantages, CASSI-based systems continue to face significant challenges related to signal reconstruction algorithms, which restrict their commercial deployment. Recent advances in deep learning, particularly Convolutional Neural Network (CNN),have introduced innovative solutions for reconstructing hyperspectral images from 2D data. However, these methods typically demand substantial computational resources, making their implementation challenging for edge devices such as smartphones and drones. In this paper, we propose a lightweight multi-scale feature extraction and mask-gated convolutional network for hyperspectral image reconstruction. The network leverages lightweight design strategies for efficiency, incorporating channel dimension reduction and compact convolutional structures like 1×1 and depthwise separable convolutions. It further enhances reconstruction accuracy with a Channel Attention Module (CAM) that adaptively reweights features while reducing parameters. Additionally, the network integrates multi-scale feature extraction and mask-gated convolutional layers, enabling high-quality reconstruction with minimal computational cost.Experimental results demonstrate that the proposed approach not only reduces computational complexity and parameter count but also achieve high reconstruction performance compared to existing methods.
{"title":"Multi-scale feature extraction mask gated network for hyperspectral image reconstruction","authors":"Xiangfeng Xie , Ping Xu , Ji Xu , Wenjie Zhang , Yian Liu , Haifeng Zheng","doi":"10.1016/j.optlaseng.2025.109543","DOIUrl":"10.1016/j.optlaseng.2025.109543","url":null,"abstract":"<div><div>Hyperspectral images (HSIs) captures more comprehensive spectral information than traditional RGB imaging, offering significant potential across various applications. Conventional methods, such as point, line, and wavelength scanning, are capable of generating hyperspectral images, but are often time-consuming and costly. Snapshot imaging systems, like the Coded Aperture Snapshot Spectral Imaging (CASSI), snap three-dimensional hyperspectral data into two-dimensional measuremen, enabling faster acquisition, reduced costs, and enhanced miniaturization. Despite these advantages, CASSI-based systems continue to face significant challenges related to signal reconstruction algorithms, which restrict their commercial deployment. Recent advances in deep learning, particularly Convolutional Neural Network (CNN),have introduced innovative solutions for reconstructing hyperspectral images from 2D data. However, these methods typically demand substantial computational resources, making their implementation challenging for edge devices such as smartphones and drones. In this paper, we propose a lightweight multi-scale feature extraction and mask-gated convolutional network for hyperspectral image reconstruction. The network leverages lightweight design strategies for efficiency, incorporating channel dimension reduction and compact convolutional structures like 1×1 and depthwise separable convolutions. It further enhances reconstruction accuracy with a Channel Attention Module (CAM) that adaptively reweights features while reducing parameters. Additionally, the network integrates multi-scale feature extraction and mask-gated convolutional layers, enabling high-quality reconstruction with minimal computational cost.Experimental results demonstrate that the proposed approach not only reduces computational complexity and parameter count but also achieve high reconstruction performance compared to existing methods.</div></div>","PeriodicalId":49719,"journal":{"name":"Optics and Lasers in Engineering","volume":"199 ","pages":"Article 109543"},"PeriodicalIF":3.7,"publicationDate":"2025-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145841609","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-24DOI: 10.1016/j.optlaseng.2025.109564
Peixin Qu , Lulu Meng , Guohou Li , Jianping Wang , Wenyi Zhao , Zheng Liang , Weidong Zhang
Accurately evaluating underwater image quality is essential in marine engineering. Light attenuation and scattering in water often result in various visual distortions, including color loss, reduced contrast, and diminished visibility. In practical applications, these distortions may have a detrimental effect on the accurate evaluation of underwater images. To address the issue, we propose a method called Multidimensional Perceptual Image Quality Evaluation (MDP-IQE), designed to accurately and efficiently evaluate the underwater image quality. This method evaluates image quality across six key dimensions: color, contrast, visibility, geometry, noise, and water quality. The Gaussian Process Regression (GPR) model is trained to relate the extracted characterizations to the subjective quality scores of the underwater images. The quality-aware characterization vectors are then extracted for each test image and input into the trained model for quality prediction. Extensive testing on two datasets, which also show great compatibility with human subjective visual perception, demonstrates the superiority of our suggested MDP-IQE method.
{"title":"Underwater image quality evaluation via multidimensional perceptual characterization","authors":"Peixin Qu , Lulu Meng , Guohou Li , Jianping Wang , Wenyi Zhao , Zheng Liang , Weidong Zhang","doi":"10.1016/j.optlaseng.2025.109564","DOIUrl":"10.1016/j.optlaseng.2025.109564","url":null,"abstract":"<div><div>Accurately evaluating underwater image quality is essential in marine engineering. Light attenuation and scattering in water often result in various visual distortions, including color loss, reduced contrast, and diminished visibility. In practical applications, these distortions may have a detrimental effect on the accurate evaluation of underwater images. To address the issue, we propose a method called Multidimensional Perceptual Image Quality Evaluation (MDP-IQE), designed to accurately and efficiently evaluate the underwater image quality. This method evaluates image quality across six key dimensions: color, contrast, visibility, geometry, noise, and water quality. The Gaussian Process Regression (GPR) model is trained to relate the extracted characterizations to the subjective quality scores of the underwater images. The quality-aware characterization vectors are then extracted for each test image and input into the trained model for quality prediction. Extensive testing on two datasets, which also show great compatibility with human subjective visual perception, demonstrates the superiority of our suggested MDP-IQE method.</div></div>","PeriodicalId":49719,"journal":{"name":"Optics and Lasers in Engineering","volume":"199 ","pages":"Article 109564"},"PeriodicalIF":3.7,"publicationDate":"2025-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145841611","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-24DOI: 10.1016/j.optlaseng.2025.109569
Fengrui Li , Zhigang Han , Jiarui Xu , Rihong Zhu
To mitigate vibration-induced phase errors in the demodulated wavefront of time-domain phase-shifting interferometry (PSI) systems, we present a synchronous calibration-acquisition method that precisely records sequential phase-shifting interferograms triggered by the phase-shifting features extracted from near-focal light signals. By integrating an electronic data processing module into the Fizeau interferometer, a closed-loop control system is established to dynamically track the peak-valley values of near-focal light intensity and generate real-time synchronization signals to the interferometer camera, enabling online phase-shift calibration and millisecond-level interferogram acquisition. This method can achieve non-isochronous sampling of four interferograms within 61 ms. Experimental results demonstrate that the method effectively suppresses ripple artifacts in demodulated wavefronts and reduces the incidence of inverted wavefronts, which can improve the operational stability of time-domain PSI systems in non-steady-state environments.
{"title":"Accurate and synchronous calibration-acquisition of phase-shifting interferograms under vibration with near-focal feature triggering","authors":"Fengrui Li , Zhigang Han , Jiarui Xu , Rihong Zhu","doi":"10.1016/j.optlaseng.2025.109569","DOIUrl":"10.1016/j.optlaseng.2025.109569","url":null,"abstract":"<div><div>To mitigate vibration-induced phase errors in the demodulated wavefront of time-domain phase-shifting interferometry (PSI) systems, we present a synchronous calibration-acquisition method that precisely records sequential phase-shifting interferograms triggered by the phase-shifting features extracted from near-focal light signals. By integrating an electronic data processing module into the Fizeau interferometer, a closed-loop control system is established to dynamically track the peak-valley values of near-focal light intensity and generate real-time synchronization signals to the interferometer camera, enabling online phase-shift calibration and millisecond-level interferogram acquisition. This method can achieve non-isochronous sampling of four interferograms within 61 ms. Experimental results demonstrate that the method effectively suppresses ripple artifacts in demodulated wavefronts and reduces the incidence of inverted wavefronts, which can improve the operational stability of time-domain PSI systems in non-steady-state environments.</div></div>","PeriodicalId":49719,"journal":{"name":"Optics and Lasers in Engineering","volume":"199 ","pages":"Article 109569"},"PeriodicalIF":3.7,"publicationDate":"2025-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842189","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-24DOI: 10.1016/j.optlaseng.2025.109558
Jinye Miao , Yingjie Shi , Fuyao Cai , Yi Wei , Lingfeng Liu , Lianfa Bai , Enlai Guo , Jing Han
Non-line-of-sight (NLOS) imaging aims at reconstructing objects around corners and is promising for diverse applications. A fundamental problem is that the imaging resolution of NLOS methods based on time-of-flight (TOF) is constrained by the single-photon timing resolution (SPTR) of the hardware’s photon counting. In this paper, a super timing-resolution method named laser pulses multiplexing (LPM) is proposed to overcome the inherent photon counting limitations of the hardware. Specifically, based on the consistency of the laser pulse response, we can utilize the photon histogram of multiple laser pulse cycles and the fixed SPTR to form a sub-single-photon timing resolution (sub-SPTR) time-difference modulation matrix. In this way, without the necessity for additional optical components or multiple exposures, a high-SPTR transient images can be decoupled through time difference modulation. Through comprehensive evaluation of the experimental data, we demonstrate that LPM enhances SPTR by at least 1 order of magnitude, enabling transient image reconstruction with a 64-picosecond single-photon timing resolution—over 10 × higher than the hardware’s intrinsic 704-ps single-photon timing resolution. Furthermore, the proposed LPM exhibits robustness against Poisson noise induced by under-scanning conditions. Especially when the scanning points are reduced to about 5 % of full samples, the structural similarity index measure (SSIM) of the reconstructed object by LPM is 0.2 higher than that without LPM. In addition, experiments show that the proposed method is also applicable to non-confocal systems, which aids in the application of array detectors. The method reduces the reliance on high-SPTR detectors using pulse multiplexing modulation, which provides a reference for combining prior modulation to overcome hardware deficiencies.
{"title":"Super-resolution non-line-of-sight imaging with laser pulses multiplexing","authors":"Jinye Miao , Yingjie Shi , Fuyao Cai , Yi Wei , Lingfeng Liu , Lianfa Bai , Enlai Guo , Jing Han","doi":"10.1016/j.optlaseng.2025.109558","DOIUrl":"10.1016/j.optlaseng.2025.109558","url":null,"abstract":"<div><div>Non-line-of-sight (NLOS) imaging aims at reconstructing objects around corners and is promising for diverse applications. A fundamental problem is that the imaging resolution of NLOS methods based on time-of-flight (TOF) is constrained by the single-photon timing resolution (SPTR) of the hardware’s photon counting. In this paper, a super timing-resolution method named laser pulses multiplexing (LPM) is proposed to overcome the inherent photon counting limitations of the hardware. Specifically, based on the consistency of the laser pulse response, we can utilize the photon histogram of multiple laser pulse cycles and the fixed SPTR to form a sub-single-photon timing resolution (sub-SPTR) time-difference modulation matrix. In this way, without the necessity for additional optical components or multiple exposures, a high-SPTR transient images can be decoupled through time difference modulation. Through comprehensive evaluation of the experimental data, we demonstrate that LPM enhances SPTR by at least 1 order of magnitude, enabling transient image reconstruction with a 64-picosecond single-photon timing resolution—over 10 × higher than the hardware’s intrinsic 704-ps single-photon timing resolution. Furthermore, the proposed LPM exhibits robustness against Poisson noise induced by under-scanning conditions. Especially when the scanning points are reduced to about 5 % of full samples, the structural similarity index measure (SSIM) of the reconstructed object by LPM is 0.2 higher than that without LPM. In addition, experiments show that the proposed method is also applicable to non-confocal systems, which aids in the application of array detectors. The method reduces the reliance on high-SPTR detectors using pulse multiplexing modulation, which provides a reference for combining prior modulation to overcome hardware deficiencies.</div></div>","PeriodicalId":49719,"journal":{"name":"Optics and Lasers in Engineering","volume":"199 ","pages":"Article 109558"},"PeriodicalIF":3.7,"publicationDate":"2025-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145841610","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-20DOI: 10.1016/j.optlaseng.2025.109562
Franzette Paz-Buclatin , Leopoldo Luis Martin , Alejandro González-Orive , Urma González-Tombolato , Kei Kamada , Akira Yoshikawa , Airán Ródenas Seguí
We report the first successful fabrication of monolithic all Yttrium Aluminum Garnet (YAG) crystalline microdisks directly in the bulk of a crystal, through a greatly simplified two-step process of three-dimensional femtosecond laser writing followed by chemical wet-etching, as compared to current multi-step approaches. The fabricated and optically characterized microdisk is 16.9 μm in diameter and 0.8 μm in thickness. We also present the first systematic study of surface tension reshaping in YAG by means of thermal annealing, identifying an optimal annealing temperature of 1475 °C for 5 h for smoothening surface irregularities. Optical characterization using tapered fiber loop evanescent coupling revealed a more than twofold improvement in the intrinsic quality factor of the Whispering Gallery Mode resonances, increasing from 3.6 × 10³ to 9.5 × 10³ after annealing. Furthermore, the YAG microdisks demonstrated outstanding thermal robustness, showing no observable morphological changes up to 1180 °C. This work establishes a robust and straightforward platform for fabricating monolithic inside-crystal YAG microresonators, enabling their application on chip-scale solid-state lasing and extreme environment sensing.
{"title":"Fabrication of a monolithic all-YAG crystalline microresonator through femtosecond laser nanolithography and thermal annealing","authors":"Franzette Paz-Buclatin , Leopoldo Luis Martin , Alejandro González-Orive , Urma González-Tombolato , Kei Kamada , Akira Yoshikawa , Airán Ródenas Seguí","doi":"10.1016/j.optlaseng.2025.109562","DOIUrl":"10.1016/j.optlaseng.2025.109562","url":null,"abstract":"<div><div>We report the first successful fabrication of monolithic all Yttrium Aluminum Garnet (YAG) crystalline microdisks directly in the bulk of a crystal, through a greatly simplified two-step process of three-dimensional femtosecond laser writing followed by chemical wet-etching, as compared to current multi-step approaches. The fabricated and optically characterized microdisk is 16.9 μm in diameter and 0.8 μm in thickness. We also present the first systematic study of surface tension reshaping in YAG by means of thermal annealing, identifying an optimal annealing temperature of 1475 °C for 5 h for smoothening surface irregularities. Optical characterization using tapered fiber loop evanescent coupling revealed a more than twofold improvement in the intrinsic quality factor of the Whispering Gallery Mode resonances, increasing from 3.6 × 10³ to 9.5 × 10³ after annealing. Furthermore, the YAG microdisks demonstrated outstanding thermal robustness, showing no observable morphological changes up to 1180 °C. This work establishes a robust and straightforward platform for fabricating monolithic inside-crystal YAG microresonators, enabling their application on chip-scale solid-state lasing and extreme environment sensing.</div></div>","PeriodicalId":49719,"journal":{"name":"Optics and Lasers in Engineering","volume":"199 ","pages":"Article 109562"},"PeriodicalIF":3.7,"publicationDate":"2025-12-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145799704","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-20DOI: 10.1016/j.optlaseng.2025.109544
Longkun He , Hao Song , Rui Liu, Xuanyu Liao, Shan Zhao, Chengjiang Zhou, Yang Yang
Infrared and visible image fusion aims to integrate information from multi-modal source images to generate more comprehensive scene representations. Although deep learning-based fusion methods have achieved significant progress, their prevalent high computational complexity and memory requirements constrain practical application potential. While knowledge distillation (KD) techniques have been introduced to mitigate this issue, existing methods typically rely solely on a single teacher model for knowledge transfer, resulting in oversimplified knowledge acquisition by student models. To address this, we propose a novel Multi-Teacher Knowledge Distillation Fusion (MTKDFusion) framework that integrates multiple pre-trained teacher models as complementary knowledge sources, fusing and enhancing their outputs to construct target fused images. For effective cross-architecture knowledge transfer, we design a target image reconstruction network based on the Laplacian pyramid as the teacher network to generate intermediate hierarchical knowledge, while developing a pseudo-twin Laplacian pyramid fusion network as the lightweight student network. Additionally, we introduce the privileged information mechanism to further enhance knowledge distillation efficiency. Compared with nine state-of-the-art methods, MTKDFusion ranks highly in the six fusion metrics (SCD, MS_SSIM, CC, QNCIE, EN, SD), and attains the highest in semantic segmentation and in object detection. Furthermore, MTKDFusion achieves practical deployment efficiency with minimal parameters (0.0728M) and fast inference (average 0.0234s across all four datasets). The source code repository is located at https://github.com/ShadowVanguard00/MTKDFusion.
{"title":"MTKDFusion: Multi-teacher knowledge distillation for infrared and visible image fusion","authors":"Longkun He , Hao Song , Rui Liu, Xuanyu Liao, Shan Zhao, Chengjiang Zhou, Yang Yang","doi":"10.1016/j.optlaseng.2025.109544","DOIUrl":"10.1016/j.optlaseng.2025.109544","url":null,"abstract":"<div><div>Infrared and visible image fusion aims to integrate information from multi-modal source images to generate more comprehensive scene representations. Although deep learning-based fusion methods have achieved significant progress, their prevalent high computational complexity and memory requirements constrain practical application potential. While knowledge distillation (KD) techniques have been introduced to mitigate this issue, existing methods typically rely solely on a single teacher model for knowledge transfer, resulting in oversimplified knowledge acquisition by student models. To address this, we propose a novel Multi-Teacher Knowledge Distillation Fusion (MTKDFusion) framework that integrates multiple pre-trained teacher models as complementary knowledge sources, fusing and enhancing their outputs to construct target fused images. For effective cross-architecture knowledge transfer, we design a target image reconstruction network based on the Laplacian pyramid as the teacher network to generate intermediate hierarchical knowledge, while developing a pseudo-twin Laplacian pyramid fusion network as the lightweight student network. Additionally, we introduce the privileged information mechanism to further enhance knowledge distillation efficiency. Compared with nine state-of-the-art methods, MTKDFusion ranks highly in the six fusion metrics (SCD, MS_SSIM, CC, QNCIE, EN, SD), and attains the highest <span><math><mi>mIoU</mi></math></span> in semantic segmentation and <span><math><msub><mi>mAP</mi><mn>50</mn></msub></math></span> in object detection. Furthermore, MTKDFusion achieves practical deployment efficiency with minimal parameters (0.0728M) and fast inference (average 0.0234s across all four datasets). The source code repository is located at <span><span>https://github.com/ShadowVanguard00/MTKDFusion</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49719,"journal":{"name":"Optics and Lasers in Engineering","volume":"199 ","pages":"Article 109544"},"PeriodicalIF":3.7,"publicationDate":"2025-12-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145799712","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Aberration control is in demand in various applications, including vision correction, improvement of imaging systems of mobile devices, optical microscopes and telescopes, optical systems for remote sensing of the Earth, and information transmission in free space. However, existing interference methods face limitations: classical approaches suffer vibration sensitivity and constrained reference beam choices; Shack-Hartmann sensors provide indirect measurement while Zernike phase contrast and correlation methods detect only weak aberrations; and adaptive techniques converge to local minima via iterative single-point point-spread-function analysis. In this work we propose self-interference, a highly sensitive vibration-resistant technique, which employs single-arm interferometric setup and allows to determine phase aberrations. To form a set of interferograms in one plane, a new class of multichannel self-referential DOEs with a tuned focus is developed, which, together with a trained neural network, it allows recording aberrations. Determination of phase distortions of the wavefront is achieved by training a convolutional neural network on 2304 synthesized patterns. To verify the calculated optical element combining the reference and the studied beams, an experiment was performed using a spatial light modulator. In the experimental implementation we studied, the DOE formed 25 self-referential orders. The main result is high-precision recognition of wavefront distortions with an absolute error of 0.0055 for model interferograms. The proposed interference technique can be used in real time using a trained neural network and is applicable to such problems as optical wireless transmission of information under turbulence, design and manufacture of optical elements.
{"title":"Self-interference defocusing multichannel diffractive optical element for neural-networks-assisted aberration recognition","authors":"P.A. Khorin , A.V. Chernykh , V.S. Shumigai , A.P. Dzyuba , S.N. Khonina , N.V. Petrov","doi":"10.1016/j.optlaseng.2025.109551","DOIUrl":"10.1016/j.optlaseng.2025.109551","url":null,"abstract":"<div><div>Aberration control is in demand in various applications, including vision correction, improvement of imaging systems of mobile devices, optical microscopes and telescopes, optical systems for remote sensing of the Earth, and information transmission in free space. However, existing interference methods face limitations: classical approaches suffer vibration sensitivity and constrained reference beam choices; Shack-Hartmann sensors provide indirect measurement while Zernike phase contrast and correlation methods detect only weak aberrations; and adaptive techniques converge to local minima via iterative single-point point-spread-function analysis. In this work we propose self-interference, a highly sensitive vibration-resistant technique, which employs single-arm interferometric setup and allows to determine phase aberrations. To form a set of interferograms in one plane, a new class of multichannel self-referential DOEs with a tuned focus is developed, which, together with a trained neural network, it allows recording aberrations. Determination of phase distortions of the wavefront is achieved by training a convolutional neural network on 2304 synthesized patterns. To verify the calculated optical element combining the reference and the studied beams, an experiment was performed using a spatial light modulator. In the experimental implementation we studied, the DOE formed 25 self-referential orders. The main result is high-precision recognition of wavefront distortions with an absolute error of 0.0055 for model interferograms. The proposed interference technique can be used in real time using a trained neural network and is applicable to such problems as optical wireless transmission of information under turbulence, design and manufacture of optical elements.</div></div>","PeriodicalId":49719,"journal":{"name":"Optics and Lasers in Engineering","volume":"199 ","pages":"Article 109551"},"PeriodicalIF":3.7,"publicationDate":"2025-12-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145799703","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-20DOI: 10.1016/j.optlaseng.2025.109565
Shanhong Ye , Feifan Yu , Zebei Mao , Jiqiang Wang
Digital Image Correlation (DIC) is an optical measurement technology widely used in engineering for the assessment of object deformation. In recent years, DIC networks based on the FlowNet framework have demonstrated promising results in measuring complex small deformations. However, these networks have limited measurement ranges, which fail to meet the demands of practical applications in engineering.
This paper proposes a novel DIC network architecture, RefDICNet, which employs a multi-stage refinement strategy, iteratively refining the displacement field across multiple scales from coarse to fine. In response to the unique challenges posed by DIC tasks, we have designed specialized components, including a dedicated backbone feature extraction layer and a multi-scale feature fusion module. Experimental results indicate that, compared to other state-of-the-art methods, RefDICNet achieves exceptional performance in predicting large displacements, small displacements, and high-frequency complex deformations, demonstrating robust performance in both synthetic datasets and real-world experiments.
{"title":"Iterative generation of accurate displacement fields across multiple scales in digital image correlation","authors":"Shanhong Ye , Feifan Yu , Zebei Mao , Jiqiang Wang","doi":"10.1016/j.optlaseng.2025.109565","DOIUrl":"10.1016/j.optlaseng.2025.109565","url":null,"abstract":"<div><div>Digital Image Correlation (DIC) is an optical measurement technology widely used in engineering for the assessment of object deformation. In recent years, DIC networks based on the FlowNet framework have demonstrated promising results in measuring complex small deformations. However, these networks have limited measurement ranges, which fail to meet the demands of practical applications in engineering.</div><div>This paper proposes a novel DIC network architecture, RefDICNet, which employs a multi-stage refinement strategy, iteratively refining the displacement field across multiple scales from coarse to fine. In response to the unique challenges posed by DIC tasks, we have designed specialized components, including a dedicated backbone feature extraction layer and a multi-scale feature fusion module. Experimental results indicate that, compared to other state-of-the-art methods, RefDICNet achieves exceptional performance in predicting large displacements, small displacements, and high-frequency complex deformations, demonstrating robust performance in both synthetic datasets and real-world experiments.</div></div>","PeriodicalId":49719,"journal":{"name":"Optics and Lasers in Engineering","volume":"199 ","pages":"Article 109565"},"PeriodicalIF":3.7,"publicationDate":"2025-12-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145799711","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号