Issue Ebookcover storyIssue CoverIssue Contents
Online FirstList of IssuesDocument Types
Spiral phase contrast microscopy for improvement of artificial intelligence cell recognition AI Introduction
Abstract：Artificial intelligence （AI）-based cell recognition in biological images is increasingly important in medical diagnostics. To improve recognition accuracy by acquiring images better suited to AI analysis， a spiral phase-contrast microscopy system was developed. A radial Hilbert transform was applied before image acquisition， and both the transmission amplitude and phase information of the specimen were used to directly generate edge-enhanced microscopic images. The generalist AI algorithm Cellpose-SAM was subsequently employed for cell contour segmentation and cell pose prediction. In proof-of-concept experiments with oral epithelial cells， spiral phase-contrast filtering substantially improved both the cell recognition rate and the accuracy of predicted cell poses. A recognition rate of 96% was achieved with spiral phase-contrast images， compared with 54% for bright-field images. The applicability of the method across cell types was further validated using red blood cells， Hepa 1-6 hepatoma cells， and human umbilical vein endothelial cells. These findings demonstrate the potential of spiral phase-contrast microscopy as a broadly applicable optical preprocessing approach for improving AI-based cell recognition accuracy and supporting algorithm generalizability across diverse biological specimens.Keywords：spiral-phase-contrast microscopy;image edge enhancement;image pre-processing;artificial intelligence;cell recognition82|19|0<HTML><L-PDF> <WORD> <Meta-XML>Updated：2026-10-20- Abstract：During in-orbit operation， the background radiation of high-sensitivity infrared detectors is modulated by complex， time-varying temperature fields， limiting radiometric calibration accuracy. Conventional correction methods based on thermal equilibrium assumptions， ground calibration， or empirical coefficients are poorly suited to dynamic， nonequilibrium orbital thermal environments. To address this limitation， a high-precision method for real-time estimation of detector background radiance is developed. In-orbit temperature telemetry and deep-space calibration observations are used to construct a modeling database through temporal alignment， outlier removal， and dataset partitioning， with coupled temperature measurements from multiple detector-associated components selected as model inputs. A random forest model is then established to capture the nonlinear relationship between these temperature fields and background radiance， overcoming the reliance of conventional methods on static assumptions and linear empirical relationships. Model training， validation， and application assessment are conducted across different satellite types and infrared spectral bands， enabling real-time background radiance estimation under complex orbital thermal conditions. Validation results yield a maximum coefficient of determination （R²） of 0.985 4， a relative bias （RBias） as low as 0.069%， and a root mean square error （RMSE） as low as 0.009 6 W/（sr·m²）. An overall mean relative error of 3.88% is achieved in typical ground-target observation scenarios. These results demonstrate that the nonlinear coupling between complex temperature fields and detector background radiation can be effectively characterized by the proposed method， providing a new approach to in-orbit background radiance estimation and improved radiometric calibration accuracy for infrared detectors.Keywords：infrared radiometric calibration;instrument background radiation;time-varying temperature field;data-driven;random forest;real-time estimation38|35|0<HTML><L-PDF> <WORD> <Meta-XML>Updated：2026-10-20
Thermal disturbance compensation for single-camera three-dimensional digital image correlation AI Introduction
Abstract：Air refractive index fluctuations in high-temperature environments can introduce spurious displacements into digital image correlation （DIC） measurements. Existing single-camera thermal disturbance compensation methods are largely restricted to two-dimensional DIC， whereas conventional dual-camera three-dimensional （3D） DIC is constrained by camera placement and extrinsic parameter stability when optical access is limited. To address these limitations， a single-camera method for 3D displacement reconstruction and thermal disturbance compensation is proposed. Left and right virtual views are generated using a single-camera， dual-prism pseudo-stereo DIC system to reconstruct 3D displacements. A dual-channel coaxial projected-speckle compensation scheme is introduced： the blue projected-speckle channel is used to characterize the thermal disturbance displacement field， while the green sprayed-speckle channel captures the combined structural displacement and thermal disturbance. Compensation is achieved by subtracting the displacement fields obtained from the two channels. Under pure thermal disturbance conditions， strong agreement is observed between the measured displacement fields， with a Pearson correlation coefficient of 0.989 870. In a rigid-body displacement experiment subjected to thermal disturbance， the root-mean-square error （RMSE） of the mean 3D deformation vector magnitude over the entire loading process is reduced from 0.074 356 to 0.028 773 mm after compensation. In a high-temperature furnace experiment， the RMSE of the x-direction displacement over the entire loading process is reduced from 0.085 303 to 0.017 240 mm. These results demonstrate the effectiveness of the proposed method in mitigating thermal disturbance errors and establish a feasible approach to non-contact， full-field 3D deformation measurement in high-temperature environments with restricted optical access.Keywords：digital image correlation;single-camera 3D measurement;thermal disturbance compensation;projected speckle;dual-channel coaxial configuration29|16|0<HTML><L-PDF> <WORD> <Meta-XML>Updated：2026-10-20- Abstract：Visible-light cameras are the most widely used optical payloads on airborne electro-optical platforms. Due to their complex and variable operating environments， the automatic focusing function is one of the key factors ensuring imaging quality. To achieve automatic focusing for visible-light cameras， an autofocus system was developed， and the key algorithms involved in the system were investigated. First， from an engineering design perspective， the working principle and overall structure of the autofocus system are systematically introduced. Subsequently， an improved image sharpness evaluation algorithm is proposed， which measures variations in image clarity by analyzing the multi-scale frequency-band energy. Finally， experiments and analyses were conducted using an electro-optical platform equipped with the proposed focusing system. Experimental results show that the autofocus system can stably achieve automatic focusing. The improved algorithm exhibits better monotonicity and a larger dynamic range， with no local peaks. The autofocus time is less than 2 s and the accuracy exceeds 98%， meeting the requirements for image clarity in both real-time tracking and post-processing on electro-optical platforms.Keywords：electro-optical platform;visible-light camera;auto focus;image sharpness evaluation;multi-scale frequency-band energy401|196|0<HTML><L-PDF> <WORD> <Meta-XML>Updated：2026-10-20
Lightweight hybrid refractive-diffractive optical system design and principal-mode PSF modeling for spatially varying restoration AI Introduction
Abstract：A lightweight hybrid refractive-diffractive optical system and an optically constrained principal-mode point spread function （OC-PM-PSF） model are proposed for small-satellite visible-light imaging payloads subject to spatially varying degradation. A Binary 2 diffractive surface is incorporated into a double-Gauss configuration， and the system is optimized across the 450~660 nm spectral range and a 34° full field of view to minimize axial chromatic aberration， improve modulation transfer function （MTF） consistency across the field， and maintain compact spot sizes. Diffractive optical element （DOE） diffraction efficiency， spectral weighting， detector response， field position， and variations in MTF and root-mean-square （RMS） spot radius are incorporated into the principal-mode decomposition of broadband PSFs. The reconstructed spatially varying PSFs are integrated with a Richardson-Lucy algorithm with total variation regularization （RL-TV） for non-blind image restoration. The lens count is reduced from six to four， the maximum clear aperture is decreased from 19.130 mm to 13.998 mm， and the estimated front-end optomechanical mass is reduced by 34.9%. At unsampled field positions， the mean and maximum relative reconstruction errors for broadband PSFs are 4.98% and 7.76%， respectively. Following restoration， PSNR is increased from 22.54 dB to 30.11 dB， and SSIM is improved from 0.840 to 0.940. Additional experiments using nonidentical degradation PSFs and mismatched restoration PSFs demonstrate robustness to limited modeling errors. These results support the integration of lightweight optical design and optically constrained PSF modeling for spatially varying image restoration in small-satellite imaging payloads.Keywords：hybrid refractive-diffractive optics;diffractive optical element;lightweight imaging;OC-PM-PSF;spatially varying restoration;RL-TV algorithm21|13|0<HTML><L-PDF> <WORD> <Meta-XML>Updated：2026-10-20Precision displacement measurement based on vortex beam interferometry： progress， challenges， and perspectives AI Introduction
Abstract：Ultraprecise optical displacement measurement is a key enabling technology for nanoscience， ultraprecision machining， and semiconductor equipment manufacturing. In this review， research advances in conjugate vortex interferometric displacement measurement are systematically examined. With a focus on the sensing paradigm based on structured orbital angular momentum （OAM） interference fields， the intrinsic advantages of this technology in enhancing detection sensitivity， extending the unambiguous measurement range， and enabling compact optical-path integration are summarized， and the principal technical bottlenecks limiting practical implementation are identified. Six major technical approaches are classified and reviewed： high-precision image and phase demodulation， optical-field optimization for enhanced angular resolution， high-speed dynamic microvibration measurement， cross-cycle extension of the unambiguous measurement range， simultaneous measurement of multiple degrees of freedom， and distributed deformation reconstruction of nonplanar surfaces. Synergistic improvements in measurement accuracy， dynamic response bandwidth， and environmental robustness may be achieved through the integration of optical metasurfaces， fiber-grating-based micro- and nanoscale modulation devices， artificial-intelligence-assisted demodulation algorithms， dual-wavelength equivalent-range extension， and comprehensive compensation for multiple error sources across the measurement domain. These developments are expected to advance the application of conjugate vortex interferometry to nanoscale positioning， precision manufacturing， curved-surface deformation measurement， and online condition monitoring of high-end equipment. This review provides a theoretical foundation and technical guidance for the development of next-generation ultraprecise optical metrology systems.Keywords：precision measurement;optical interference;vortex beams;orbital angular momentum;displacement measurement19|13|0<HTML><L-PDF> <WORD> <Meta-XML>Updated：2026-10-20
Modern Applied Optics
High-speed dynamic focus tracking and error compensation for wafer laser stealth dicing AI Introduction
Abstract：Conventional fixed-focus wafer laser stealth dicing is susceptible to stage-motion disturbances， structural resonance， and environmental microvibrations during high-speed operation. These disturbances can induce dynamic focal displacement and time-varying defocus errors， resulting in nonuniform cutting depths， reduced processing consistency， and lower product yield. To address these limitations， a laser-triangulation-based method for dynamic focus tracking and error compensation is proposed. A real-time focal-offset detection system is developed using noncontact laser triangulation to measure dynamic focal displacement and characterize disturbances during high-speed processing. The coupled transmission mechanisms of multiple disturbance sources are analyzed， and a dynamic defocus error model is established to elucidate the relationship between these disturbances and focal displacement. A two-level compensation architecture integrating offline static calibration with online dynamic correction is adopted. A composite feedforward–feedback control algorithm incorporating peak-filter-based suppression of high-frequency disturbances is further designed to enable dynamic focus tracking and closed-loop defocus error compensation. Experimental results demonstrate effective suppression of dynamic defocus errors and processing disturbances. The trajectory tracking error is reduced from ±3.5 μm to ±0.50 μm， and a dynamic focal-position compensation accuracy of ±0.25 μm is achieved. Processing stability and modified-layer consistency are substantially improved， supporting the industrial application of high-speed， high-precision laser stealth dicing for high-end ultrathin wafers.Keywords：wafer laser stealth dicing;dynamic focus tracking;dynamic defocus error compensation;laser triangulation measurement25|23|0<HTML><L-PDF> <WORD> <Meta-XML>Updated：2026-10-20Dual LaNiO3 strain regulation for low-temperature PZT films and MEMS micromirror application AI Introduction
Abstract：Achieving both strong crystallographic orientation and low defect density in PZT thin films fabricated at low temperatures remains a critical challenge for the CMOS-compatible integration of piezoelectric MEMS scanning micromirrors. To address this limitation， a synergistic strain-engineering strategy employing dual LaNiO₃ （LNO） buffer layers is proposed. A Pt/LNO/PZT/LNO/Pt multilayer heterostructure is deposited on SiO₂/Si substrates by magnetron sputtering at temperatures not exceeding 500 °C. In-plane tensile strain is introduced by the bottom LNO layer to promote （001）-oriented nucleation， whereas in-plane compressive strain is imposed by the top LNO layer to suppress lead volatilization and oxygen vacancy migration. Crystallographic orientation， strain distribution， morphology， and interfacial chemistry are systematically characterized using X-ray diffraction， scanning electron microscopy， atomic force microscopy， and X-ray photoelectron spectroscopy （XPS）， complemented by electrical measurements. The optimized films are subsequently integrated into a piezoelectric MEMS scanning micromirror for device-level validation， achieving a ratio of 0.72 and a （002） rocking-curve full width at half maximum of 1.43°. An approximately one-order-of-magnitude reduction in overall defect density is corroborated by XPS and electrical measurements. The films exhibit a transverse piezoelectric coefficient e31，f of （-15.5±0.5） C/m²， a remanent polarization Pr of （23±2） μC/cm²， a relative dielectric constant εr of （1 200±50） at 1 kHz， a leakage current density of 8.3×10⁻⁷ A/cm² at 100 kV/cm， and a breakdown voltage of 80 V. An optical scanning angle of ±40° is achieved by the MEMS micromirror at a driving voltage of 25 Vpp. By enabling highly oriented， low-defect PZT films within a CMOS-compatible thermal budget， the proposed process provides a viable route to the monolithic integration of high-performance piezoelectric thin films with silicon-based MEMS.Keywords：PZT film;dual LNO buffer layers;low-temperature growth;strain engineering;MEMS micromirror18|15|0<HTML><L-PDF> <WORD> <Meta-XML>Updated：2026-10-20Gyro bias estimation for rotating inertial gyroscope using rotational angle increments observation AI Introduction
Abstract：To accelerate the convergence of vertical gyroscope bias estimation during the initial alignment of a single-axis rotational inertial navigation system （SRINS）， a method incorporating angular increment measurements from the indexing mechanism is proposed. High-precision angular increments are incorporated into a Kalman filtering framework to construct a novel observation model. The effects of gyroscope scale factor errors， misalignment errors， and indexing mechanism output errors on bias convergence are systematically analyzed. Based on this analysis， an appropriate indexing strategy and a corresponding filter model are designed to mitigate these error sources and improve robustness in practical applications. Experimental results demonstrate that a vertical gyroscope bias estimation accuracy better than 6.2 × 10⁻⁴ （°）/h is achieved within 15 min. Compared with the conventional method using velocity errors as observations， the maximum position error is reduced from 1.23 to 0.15 nautical miles in a 12-hour static navigation test and from 1.27 to 0.76 nautical miles in a vehicle-mounted navigation test. These results confirm the effectiveness of the proposed method in estimating vertical gyroscope bias and improving SRINS navigation accuracy.Keywords：single-axis rotational inertial navigation system;gyro bias estimation;initial alignment;indexing mechanism angle increment20|14|0<HTML><L-PDF> <WORD> <Meta-XML>Updated：2026-10-20
Micro\/Nano Technology and Fine Mechanics
Lightweight underwater image enhancement network with parallel fusion of axial convolution and channel attention AI Introduction
Abstract：Underwater images are degraded by wavelength-dependent absorption and scattering， resulting in color casts and reduced contrast. To address these degradations while balancing enhancement quality and computational efficiency， a lightweight underwater image enhancement network based on the parallel fusion of axial convolution and channel attention is proposed. An encoder–decoder architecture with residual compensation is adopted， with ShuffleNet V2 employed as the lightweight feature extraction backbone. At the bottleneck， axial depthwise convolution and channel attention are integrated in parallel to expand the spatial receptive field and adaptively recalibrate channel-wise feature weights， respectively. The enhanced image is reconstructed through a global residual connection. The network is trained on the LSUI dataset using a combined mean squared error （MSE） and structural similarity （SSIM） loss. Evaluations on multiple public datasets， including LSUI， UIEB， and EUVP， demonstrate superior performance over mainstream comparison methods in key metrics， including PSNR， SSIM， and UCIQE. On the LSUI test set， a PSNR of 25.13 dB and an SSIM of 0.849 1 are achieved. Without GPU acceleration， per-image inference latency is substantially lower than that of comparable lightweight methods， while a processing speed of 144.61 frame/s is achieved on the Jetson Orin NX. With only 1.60 M parameters， the proposed network delivers improved enhancement quality and computational efficiency， supporting real-time， high-quality underwater image enhancement on resource-constrained edge platforms.Keywords：underwater image enhancement;lightweight network;ShuffleNet V2;axial depthwise convolution;channel attention28|13|0<HTML><L-PDF> <WORD> <Meta-XML>Updated：2026-10-20Misaligned visible-infrared object detection based on feature compensation and enhancement fusion AI Introduction
Abstract：Visible-infrared object detection offers substantial advantages in low-light conditions and challenging environments. However， most existing methods rely on precise image registration， limiting their applicability to the cross-modal spatial misalignment commonly encountered in practice. To overcome this limitation， CEFNet， a lightweight detection framework for misaligned visible–infrared images， is proposed without requiring explicit pixel-level alignment. First， a frequency-domain decoupled feature compensation module is introduced， in which features are decomposed using the discrete wavelet transform. Cross-modal high-frequency discrepancies are then employed as soft gating signals for a state-space model to suppress the propagation of local misalignment noise during global sequential modeling， enabling efficient， noise-robust feature compensation. Second， a multibranch asymmetric convolution enhancement module is designed to expand the receptive field and strengthen modality-specific feature representations. Finally， a difference-guided fusion module is developed to adaptively suppress shared， redundant background information and emphasize discriminative complementary features. Experimental results demonstrate that CEFNet achieves an mAP50 of 87.1% and an mAP50-95 of 53.1% on the misaligned DVTOD dataset. Compared with the state-of-the-art CDC-YOLOFusion framework， the parameter count and computational complexity are reduced by 90% and 79%， respectively， while detection accuracy is improved by 0.6%-1.0%. Further evaluations on M³FD and LLVIP demonstrate enhanced robustness and generalization， supporting the applicability of CEFNet to efficient multimodal object detection under cross-modal spatial misalignment.Keywords：visible-infrared object detection;spatial misalignment;feature compensation;feature enhancement;feature fusion19|12|0<HTML><L-PDF> <WORD> <Meta-XML>Updated：2026-10-20- Abstract：Power tower detection in high-resolution satellite remote sensing imagery is hindered by substantial variations in tower scale， weak feature representations of small distribution towers， complex background interference， and limited bounding-box localization accuracy. To address these challenges， SA-DINO， a scale-aware power tower detection method based on Grounding DINO， is proposed. First， a detail-preserving small-object feature branch is designed to enhance local edges， high-frequency textures， and contextual information in high-resolution feature maps before they are passed to the Transformer encoder， mitigating the degradation of small distribution tower features during deep feature propagation. Second， a scale-aware， language-guided query selection module is introduced to adaptively allocate queries according to the spatial resolution of feature maps at different scales. Small-object candidates in high-resolution feature maps are thus given greater access to the decoder， while global high-confidence queries are retained to preserve detection performance for large transmission towers. Finally， a scale-aware small-object localization loss is formulated using the normalized Wasserstein distance to quantify geometric discrepancies between predicted and ground-truth bounding boxes. The strength of localization supervision is adaptively adjusted according to object scale to improve the stability of small-object bounding-box regression. Experiments on the SRSPTD dataset demonstrate that SA-DINO achieves an mAP50 of 80.9% and an mAP50-95 of 42.2%， exceeding the baseline by 3.5% and 1.7%， respectively， with only a 0.94% increase in parameter count. These results demonstrate that scale-aware feature enhancement， query selection， and localization supervision can effectively improve power tower detection in complex remote sensing scenarios with minimal additional model parameters.Keywords：satellite remote sensing;tower detection;Grounding DINO;scale awareness;small object33|18|0<HTML><L-PDF> <WORD> <Meta-XML>Updated：2026-10-20
Information Sciences
