Jiawei Chen, Yang Yang, Chao Yu, Yu Tian, Zhi Cao, Xue Yang, Linghao Li, Hang Su, Zhaoxia Yin. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Autistic people often experience heightened emotional and sensory demands that affect wellbeing. Creative drawing supports expression and emotion regulation, yet an open-ended process can introduce uncertainty and overstimulation, particularly for autistic individuals who prefer structure. We present the Structured Creativity Toolkit (SCT), a set of tangible artifacts including custom laser-cut stencils, composition templates, and guided color palettes, designed to scaffold creative expression and self-regulation. We first conducted a qualitative study to identify design opportunities and user needs, then evaluated SCT through a mixed-methods approach combining a controlled study with post-study interviews to examine engagement, user experience, and self-reported wellbeing. Our findings indicate that SCT increased drawing engagement and perceived drawing quality while supporting self-reported psychological and emotional benefits. This paper contributes: (1) a creativity-support toolkit for structured drawing, and (2) evidence that order-affirming scaffolds can translate preferences for structure into design resources that support autistic individuals.

Driven by the flexibility, reliability, and scalability of microservice architecture, an increasing number of enterprises are decomposing monolithic applications into microservices. However, existing deep learning-based decomposition methods rely heavily on partition number selection, which, if unreasonable, can lead to frequent microservice communication and reduced performance. Moreover, manual partition suggestions not only decrease automation but also fail to adapt to rapid business iteration, and existing methods inadequately capture the relationship characteristics between application classes. To address these issues, this paper proposes a novel graph-based partitioning technique, GGRME. It constructs a system dependency graph through static, dynamic, and semantic analysis, and then employs a self-supervised gated graph neural network combined with cross-supervised optimization for community detection to automatically generate microservice decomposition results. Experiments demonstrate that GGRME outperforms benchmark methods in 65 % of tests, yielding superior microservice decomposition performance.

Ground-based distributed radar is a potential technique for space target detection. However, in the case of low signal-to-noise ratio (SNR), it is difficult to achieve long-term integration due to the limited ephemeris guidance accuracy and complex motion model. To solve this problem, a space target detection algorithm based on digital beamforming (DBF) and generalized Radon-Fourier transform (GRFT) is proposed in this paper. To avoid the gain loss caused by ephemeris errors, small-scale beam-searching is conducted through DBF technique, which also enables the measurement of target angle and even angular velocity. Besides, transforming the problem of energy accumulation into parameterized model matching, the GRFT process can achieve long-term integration effectively in the case of complex motion models. The effectiveness of the algorithm is verified via real data experiments based on a ground-based distributed radar. By showing an effective 30-second integration and a computational efficiency improvement of 40%, the validation of the proposed algorithm has been proved.

Multistatic unmanned aerial vehicle synthetic aperture radar (MUAV-SAR) three-dimensional (3-D) imaging system suffers from the time and phase synchronization errors among multiple stations. The classical two-way direct wave pulse exchange synchronization method introduces the π-ambiguity phase error and causes limited time-phase synchronization accuracy with multiple system nodes, leading to 3-D images defocusing. An MUAV-SAR joint synchronization method based on multiple direct wave pulses exchange is proposed to solve the π-ambiguity problem robustly and improve the synchronization accuracy significantly. Firstly, the π-ambiguity phase error is estimated through the comparative calculation of delay-phase information extracted from direct wave pulses and the high estimated success probability (99.73%) of the π-ambiguity can be achieved through the noise smoothing. Secondly, the synchronization accuracy is improved, that is, the time and phase error are reduced to about √2/Nof the existing method by utilizing N stations information fusion to jointly process redundant information of direct wave pulses from multiple synchronization links. Finally, a four-station UAV SAR real data experiment verifies the effectiveness of the proposed approach.

Earth-based radar is a pivotal instrument in deep space exploration to obtain radar images of desired celestial bodies. However, the system performance and image resolution of conventional integrated Earth-based radars with only one radar are limited by the power-aperture product and cannot meet the higher demands of deep space exploration. Distributed coherent radar is a new radar system composed of multiple radar units and a central control system, and its system performance can be further improved by increasing the number of radar units. Distributed coherent radar provides a reliable way to build a high-performance and high-resolution Earth-based deep space exploration system. This article introduces several key technologies about distributed coherent radar astronomical imaging: 1) high-precision coherence parameter estimation, which ensures full-coherence performance of the distributed coherent radar; 2) high-precision nonideal effect compensation, which eliminates the image offset and defocusing induced by the nonideal effects; 3) fast factorization backprojection (FFBP) algorithm, which achieves high-resolution fast imaging of celestial bodies. Moreover, based on a distributed coherent radar prototype system composed of four radar units with an antenna aperture of 16 m, high-resolution imaging experiments of the moon are conducted, and the effectiveness of the distributed coherent radar is successfully validated, which could not only provide a reference for the distributed coherent radar system but also provide a reliable solution for detection and imaging of other celestial bodies in the solar system in the future.

The accuracy of the traditional compressed sensing (CS) based tomographic synthetic aperture radar (TomoSAR) imaging is limited by the inappropriate grid partitioning. The atomic norm based processing effectively solves this problem by implementing variable estimation in the continuous domain, that is, avoiding the undesired grid partitioning manipulation. Nevertheless, the performance of the atomic norm based TomoSAR imaging is limited in two main aspects: limited geometry adaptability caused by the uniform sampling requirement and the high computational load. In this paper, a novel atomic norm minimization (ANM) based off-grid TomoSAR imaging is proposed for the fast processing with nonuniform sampling. The main technical contributions are twofold: First, the nonuniformly sampled data is resampled to be uniform where a new geometrical projection-based interpolation is used; Second, the ANM problem is solved by using the non-symmetric cone model to speed up the processing, reducing the computational load fromO(N2) toO(N). The proposed approaches have been verified by the computer simulations and the real data experiments.

Unmanned-aerial-vehicle-based (UAV-based) ultrawideband and ultrawidebeam (UWB) synthetic aperture radar (SAR) is very sensitive to atmospheric turbulence and suffers from serious 2-D space-variant motion errors (SVMEs) caused by the ultrawide beam and frequency-dependent phase errors caused by the ultrawideband. This article proposes an autofocus approach for UAV-based UWB SAR data based on the quasi-polar grid fast factorized backprojection (FFBP) imaging framework, multiple subband local autofocus (MSBLA), and trajectory deviation estimation. First, based on an improved weighted phase gradient autofocus (WPGA) method for subband-division local images, MSBLA is introduced to solve the local motion error estimation problem with frequency-dependent phase errors. Then, trajectory deviation estimation based on the weighted least square (WLS) method is performed to solve the 2-D SVME problem. Finally, the subaperture trajectory deviations are fused into a full-aperture trajectory deviation by an improved fusion strategy based on piecewise weighting. This approach is applied to real data from a new UAV-based UWB SAR. The results of both simulation and real data experiments are presented and verify the effectiveness of the proposed approach.

Linear array (LA) multiple-input–multiple-output (MIMO) synthetic aperture radar (SAR) has the capacity of achieving 3-D images in a single pass. If processed by matched filtering-based linear 3-D imaging, sidelobe suppression is often required for image quality enhancement. However, in the case of imaging a large target in a short range, sidelobes of the target will become spatially variant and curved, and traditional sidelobe suppression methods will fail. This article proposes a new spatially variant curved sidelobe suppression method for LA MIMO SAR short-range 3-D imaging. The key technique is the employment of a new pseudopolar coordinate system where both the spatial variance and the curvature of 3-D sidelobes can be removed. Specifically, a new 3-D spatially variant apodization (SVA) kernel is applied for sidelobe suppression to maintain the resolution performance. The validity of the presented approach has been demonstrated via computer simulations, the tower crane experiment, and the unmanned ground vehicle experiment.

Translational motion compensation plays an important role in inverse synthetic aperture radar (ISAR) imaging. However, existing translational motion compensation algorithms cannot work well when the signal-to-noise ratio (SNR) is low and the target has complex motion at the same time, as the algorithms usually assume that these two situations do not occur simultaneously. To address this problem, an improved parametric translational motion compensation algorithm based on signal phase order reduction (SPOR) and minimum entropy is proposed. The key is to decrease the phase order of the signal, which has a nonlinear phase and corresponds to the complex motion, and then obtain the signal with a linear phase corresponding to the noncomplex motion. Subsequently, the signal is transformed into the Doppler domain to generate the SPOR result. Obviously, when the translational motion is well compensated, the SPOR result will be coherently accumulated and has the best quality, which means that the SPOR result has good robustness against the low SNR. Thus, the translational motion is modeled as a polynomial model, the entropy of the SPOR result is taken as the optimizing target, and the relationship between the translational motion compensation parameters and the entropy is established. Finally, coarse search and particle swarm optimization (PSO) are sequentially performed to optimize the entropy of the SPOR result and estimate the translational motion compensation parameters accurately and efficiently. Computer simulation results and experimental results based on unmanned aerial vehicle (UAV) radar validate the proposed algorithm.

Due to the extremely high orbital height and long synthetic aperture time, the geosynchronous synthetic aperture radar (GEO SAR) will inevitably suffer from different types of undesired errors, including atmosphere, orbital measurement error, antenna vibration, and scenery height fluctuation; moreover, because of the extremely large imaging swath, these undesired errors also have severe 2-D spatial variance. Thus, the autofocus processing plays a very important role in GEO SAR. However, current autofocus algorithms cannot handle all of the aforementioned complicated and 2-D spatial-variant errors simultaneously. In this article, an autofocus back projection (BP) method for GEO SAR based on minimum entropy is proposed. First, the BP algorithm based on a digital elevation model (DEM) is adopted to deal with the scenery height fluctuation. Then, the 2-D image segmentation is conducted to solve the spatial variance of the undesired errors. Subsequently, without the assumption of error type and considering both the amplitude error and phase error, the autofocus processing based on minimum entropy and adaptive moment estimation (Adam) is conducted to estimate the undesired errors iteratively and precisely. Moreover, the aperture division and sub-aperture fusion will also be utilized to alleviate the image quality degradation or even defocus, which could also improve the precision of error estimation. Finally, computer simulation results validate the effectiveness of the proposed method.

Linear-array multiple-input–multiple-output (LA-MIMO) synthetic aperture radar (SAR) can obtain 3-D radar images by only one pass. However, it is sensitive to time-variant measurement errors of curved track and time-variant attitude angles, meaning that autofocus processing for the LA-MIMO SAR tomography is necessary. The existing autofocus methods cannot be used to estimate thetime-variantand3-D space-variantmotion errors (3-D SVME) of the LA-MIMO SAR. To solve this problem, a new autofocus approach based on multiple local autofocusing and the LA-MIMO SAR time-variant motion error estimation is proposed. First, the local motion error estimation based on the fast local spectral analysis (SPECAN) 3-D imaging and the maximum contrast optimization 2-D local autofocusing is performed to estimate the local time-variant motion errors. Then, based on the linear-array motion error model, the time-variant 3-D trajectory deviations of the array center and attitude angles are estimated by the weighted least square estimation (WLSE) to solve the 3-D SVMEs. Last, the 3-D fast factorized backprojection (FFBP) is performed to obtain the well-focused 3-D image of the whole beam. The proposed approach has been applied for the tomography of a new crawler-type unmanned-ground-vehicle (UGV) LA-MIMO SAR. Both the simulation and real data experiments verify the effectiveness of the proposed approach.

A distributed configuration is a promising realization of tomographic synthetic aperture radar (TomoSAR) 3-D imaging. It is able to implement single-pass tomographic imaging in a very short time and, hence, outperforms the traditional time-consuming multipass TomoSAR. It also outperforms the traditional single-platform TomoSAR by achieving a higher resolution in elevation by forming a much larger spatial baseline. However, there has been little research reported on the distributed TomoSAR so far. In this article, we, for the first time, experimentally demonstrate the great potential of the single-pass distributed TomoSAR 3-D imaging. The main contributions are threefold. First, a new distributed TomoSAR 3-D imaging model is built, characterized by using both inner monostatic and bistatic spatial configurations. Second, a new multistatic synchronization scheme is developed for accurately correcting both multistatic time and phase synchronization errors. Finally, a P-band distributed unmanned-aerial-vehicle (UAV) TomoSAR prototype with four separate stations is built with an elaborately designed time-division waveform for full data acquisition. To the best of our knowledge, this is the first distributed TomoSAR system. We have also implemented the first single-pass TomoSAR 3-D imaging experiment and successfully achieved a meter-level 3-D image in Pinggu, Beijing, China.

The extremely-high-squint spotlight synthetic aperture radar (SAR) has been widely used in military and commercial fields. Due to the extremely-high-squint angle, the point spread function (PSF) used for SAR resolution evaluation suffers from severe distortion and thus results in non-uniform spatial resolution. To solve this problems, this paper proposes a novel radar parameter adjustment (RPA) method to correct the PSF distortion during the data acquisition. The mechanism is to achieve an orthogonal PSF by correcting the wavenumber spectrum distortion. In this case, the radar parameters, i.e., the center frequency and the chirp rate, vary at every azimuth sampling position. The proposed approach is validated via the computer simulations.

Linear array multiple-input multiple-output synthetic aperture radar (LA-MIMO SAR) tomography can provide 3-D radar images without layover and geometric distortion effects. However, suffering from the channel and motion error, and large computation, the real LA-MIMO-SAR data is difficult to be well focused quickly. Therefore, the 3-D imaging results of real LA-MIMO-SAR data are rarely shown in existing literature. In this paper, real LA-MIMO-SAR data imaging processing method based on 3-D fast factorized Backprojection is proposed to well focus the echoes with channel and motion errors, and large size of data. Then, our experiments of LA-MIMO-SAR 3-D imaging are demonstrated. The LA-MIMO radar is installed to a special tower crane and works in downward-looking and forward-looking geometry. Some 3-D imaging results are demonstrated to verify the downward-looking and forward-looking LA-MIMO-SAR modes and the processing method.

Synthetic Aperture Radar (SAR) tomography is an important technique for target elevation information inversion and reconstructs the 3D structure of the target via multi-pass observations. At present, SAR tomography is mainly used in large-scale, low-resolution scenes where the range between the radar and the target is far larger than the target size and the resolution only meters. Therefore, the high-order phase residue is small and will not affect the elevation recovery. However, in the case of the small-scale, high-resolution scenes, the traditional TomoSAR processing will introduce quadratic phase residuals and affecting the quality of recovery. In order to solve the problem, this paper proposes a new method to reconstruct the 3-D structure in high-resolution scenes via elevation segmentation recovery. The main idea is to establish a reference signal at different height sections of the target area so that the quadratic phase residual is less than π/2. Finally, the estimated result of each section is projectively transformed into a unified coordinate system to achieve stable and accurate recovery. Besides, the algorithm has been verified by simulation data and measured data.

Geosynchronous synthetic aperture radar (GEO SAR)-related technologies are being mature, and the first GEO SAR satellite is expected to launch in the next ten years. Under this circumstance, some equivalent experiments should be conducted at the current stage to validate some key characteristics or parameters, which could significantly increase the success possibility of the GEO SAR project. To validate the feasibility of GEO SAR imaging with long integration time, which is the most important and fundamental characteristic of GEO SAR, the first helicopter platform-based equivalent GEO SAR experiment with long integration time was performed in Qianxi County of China on May 22, 2019. The integration time of it is 80 s, which is carefully designed to maintain the consistence between itself and the integration time of the GEO SAR. Furthermore, the azimuth signal-to-noise ratio gain with long synthetic aperture time is analyzed. Moreover, the 2-D space-variant motion error introduced by the complex helicopter trajectory and the performances of different imaging algorithms are analyzed to choose the proper imaging algorithms; to overcome the flaws and unclarities of existing algorithms, some improvements are proposed to obtain the well-focused SAR image. What is more, the equivalence of this experiment is also analyzed detailedly to demonstrate the effectiveness of this experiment. At last, the imaging result with synthetic aperture time of 100 s and the comparison between itself and the optic photograph validate the success of this equivalent experiment and the feasibility of GEO SAR imaging with long integration time.

Image coregistration is a key step in tomographic SAR (TomoSAR) signal processing, and the quality of coregistration directly determines the tomographic results. Traditional coregistration is performed based on the far-field assumption, where the scattering characteristics remain constant with different look angles and distances. However, for near-field TomoSAR observation, the traditional coregistration method will fail because of the mismatch caused by the change of scattering characteristics. In this study, we propose a new structure-based coregistration method for near-field ground-based MIMO TomoSAR. Corse coregistration based on the structure and fine coregistration combining correlation function and singular modification are conducted to achieve coregistration accurately and robustly. The validity of the presented approach is validated by real data.