The advancement in sensitivity and field of view of next-generation wide-field survey telescopes requires astrometric measurements with high precision, even in the presence of significant geometric distortions. To address this challenge, we develop a Weighted Polynomial Distortion Correction in 2-Phase (WPDC-2P) method. This approach enhances stellar cross matching, incorporates distance-based weighting into the traditional polynomial fitting, and employs a look-up table to absorb the remaining distortion residuals. Validated on simulated data from the Main Survey Camera of the Chinese Space Station Survey Telescope (CSST), incorporating geometric distortions up to approximately 200 pixels, the method achieves astrometric standard deviation ranging from 0.013 to 0.107 pixels (0.03 pixels for the g-1 detector) across all 18 detectors. Under extreme crowding conditions (e.g., globular cluster NGC 2298), the astrometric precision for the g-1 detector reaches 0.05 pixel level within the central region (r(d) < 4000), despite a centroiding precision of similar to 0.04 pixels. When applied to the Beijing-Arizona Sky Survey data, for which the standard pipeline delivers an astrometric uncertainty of similar to 20 mas, our method reduces the positional scatter to sigma(Delta alpha) = 5.494 mas (0.01 pixels) and sigma(Delta delta) = 9.981 mas (0.02 pixels) using only a weighted third-order polynomial correction. The method has been integrated into the CSST data processing pipeline and is prepared for further refinement using on-orbit calibration data.

The conversion between visibilities and images is a fundamental yet computationally expensive operation in radio interferometric imaging. Although existing algorithms combining high-precision convolution kernels with w-stacking have achieved imaging precision on the order of 10-12, the efficient processing of massive datasets from next-generation arrays such as the SKA telescope remains a formidable challenge, constrained by the hardware limitations of heterogeneous platforms. In this work, we introduce the baseline separation paradigm (BSP), an innovative imaging framework designed to optimize workload distribution. The core strategy of BSP is to partition the dataset into two regimes: utilizing the GPU for the massive volume of short-baseline data to bypass the memory bottleneck, while leveraging the CPU for long baselines to preserve high-resolution details, with both regimes utilizing grids strictly minimized to their respective spatial frequency limits. This strategy effectively resolves the conflict between the limited memory capacity of accelerators and the large grid sizes required for wide-field imaging. We implemented a prototype to evaluate the algorithm performance. Experimental results show that, for SKA1-Low-scale simulations, BSP achieves a precision of 10-11, comparable to ducc0.wgridder, while providing a substantial speedup. This approach provides a scalable solution for future exascale astronomical data processing.

FAST pulsar surveys generate candidate volumes that make manual review a practical bottleneck, so fixed-threshold classification is often mismatched to deployment under limited review budgets. We formulate candidate screening as a constrained operating-point selection problem: given model scores, the decision threshold is chosen to maximise recall while satisfying explicit review-oriented constraints. Each candidate is represented by four diagnostic regions of interest together with a compact one-dimensional feature vector derived from curve statistics, and several representative scoring models are evaluated within a unified 2D + 1D framework. Thresholds are selected on a calibration split under either a minimum precision constraint or a maximum false-positive-rate constraint. Evaluation includes both conventional point metrics and budget-aware measures, namely, Top-K, Top-x% and workload-recall curves. Experiments on labelled FAST data show that moderate constrained operating points can provide more practical deployment trade-offs than a fixed threshold in the evaluated review-limited setting, particularly by balancing false-alarm control against pulsar recovery. Budget-aware evaluation also reveals practical differences among scoring variants that are less visible in fixed-threshold summaries. These results support treating threshold selection as an explicit deployment component rather than a static postprocessing step in FAST candidate-screening pipelines.

Solar radio bursts at very low frequencies are key phenomena in the Sun-Earth space environment, providing crucial diagnostics of the acceleration and propagation of solar wind, coronal mass ejection (CME), and non-thermal energetic particles and serving as important indicators for space weather forecasting. To meet the demand for rapid screening of burst events in large-scale observational datasets, we present an end-to-end automatic detection and evaluation framework tailored for Type III bursts, built upon long-term radio dynamic spectra from STEREO-A/SWAVES. We formulate radio burst detection as a one-dimensional interval localization task along the time axis and, in view of the scarcity of annotated samples, cast it as a few-shot object detection task. Building upon the Faster R-CNN architecture with a ResNet50-FPN backbone, we propose the Meta-FSOD framework, which adopts an episodic training paradigm to construct support-query episode pairs. The framework incorporates a metric-guided prototype learning branch to semantically align and calibrate region-of-interest (RoI) features via class prototypes, and integrates a dynamic Beta-Gating mechanism coupled with Soft-NMS to effectively suppress false positives while preserving high-recall performance. Experimental results demonstrate that, despite being trained on a significantly smaller dataset than comparable studies, Meta-FSOD achieves competitive performance, closely matching that of conventional supervised model. The proposed framework exhibits strong cross-temporal generalization capabilities and holds considerable potential for engineering applications in deep space exploration missions.

The sensitivity of nanohertz gravitational wave detection relies heavily on the data quality of pulsar timing arrays. However, observational data inevitably contain non-Gaussian outliers, which severely degrade the accuracy of parameter estimation under standard Gaussian assumptions. To address the computational complexity and parameter degeneracy of existing mixture models, this paper proposes a robust Bayesian estimation method using a Student-t likelihood. This approach leverages the heavy-tailed nature of the Student-t distribution to adaptively down-weight outliers. It achieves this without requiring complex predefined physical models or numerous latent variables. Using the temporal separation between high-frequency transient anomalies and low-frequency correlated noise, our method stabilizes the covariance at the measurement level. This effectively prevents high-frequency data contamination from bleeding into low-frequency spectra, a common issue in traditional joint fitting. Simulations with controlled anomaly injections demonstrated the method's high precision and recall. We then applied the model to the NANOGrav 9-year dataset, specifically observing the millisecond pulsar PSR J1909-3744. For the 800-GASP backend data, known for significant non-Gaussian noise, the model's degrees of freedom converged to 5.2. In contrast, in the uncontaminated 800-GUPPI backend data, the degrees of freedom naturally converged toward the Gaussian limit of 23.6. Additional validation on PSR B1937+21 further demonstrates that the proposed weighting mechanism remains robust in the presence of significant intrinsic red noise. This confirms the safety and remarkably low false-positive rate of our adaptive down-weighting mechanism. Ultimately, this mathematically straightforward and highly generalizable method provides a more reliable statistical foundation for precision timing and gravitational wave searches in complex noise environments.

Improving the accuracy of photometric redshifts (photo-z) is essential for reliable statistical studies of cosmology and galaxy evolution. However, missing photometric bands are a common observational challenge that can significantly degrade photo-z estimation accuracy. In this work, we present a systematic evaluation of data imputation methods aimed at improving photo-z performance. We benchmark a range of representative machine learning and deep learning architectures, identifying k-nearest neighbors (KNN) and the attentionbased SAITS model as the leading performers. These models are then applied to China Space Station Survey Telescope mock data to assess their performance under realistic observational conditions. Our results show that KNN yields the highest accuracy under idealized missing completely at random (MCAR) conditions with complete training sets, whereas robustness tests reveal that SAITS significantly outperforms KNN when training data are incomplete or when applied to realistic mixed-mechanism scenarios. We find that domain consistency between training and testing missingness patterns is a prerequisite for optimal performance, highlighting the risks of domain shift in supervised regression tasks. Furthermore, our analysis demonstrates that while general imputation models are highly effective for MCAR and missing at random data, they are detrimental when applied to missing not at random data arising from flux limits, as statistical models fail to capture the physical information inherent in these nondetection. Consequently, we advocate for more sophisticated architectures capable of disentangling stochastic missingness from physical nondetection to address these distinct mechanisms individually.

High-energy charged particles in cosmic ray (CR) generate anomalous signals or noise artifacts when colliding with astronomical detectors, introducing distortions in both imaging and spectral data. This phenomenon poses a significant challenge in differentiating celestial signatures from cosmic ray-induced artifacts, particularly during observation scenarios involving resolved galaxies. In this study, we propose a deep learning framework that integrates the multi-scale attention mechanism and dynamic adaptive loss function for CR detection. The multi-scale linear attention mechanism is adopted to achieve a synergistic perception of global context modelling and local texture features in resolved galaxies. The large kernel selection block is used to effectively extend the capture range of global dependence of CR structures. The efficient multi-scale attention block is introduced to further enhance the texture differentiation between CR and celestial fringes of resolved galaxies. Furthermore, a dynamic weighted loss function based on a Gaussian residual response is introduced, with the aim of adjusting the negative sample gradient weights through statistical analysis of background noise patterns adaptively. This approach significantly improves model performance in class-imbalanced CR detection tasks. Experimental results demonstrate that the proposed model achieves consistent performance improvements on enhancements in CR detection for resolved galaxies.

Next-generation radio interferometers, with their enhanced spatial, temporal, and high-frequency resolution, will pose considerable challenges to data processing and storage. Frequency averaging reduces the volume of data, but would cause smearing of the bandwidth. The examination of bandwidth smearing is essential in the processing of radio interferometer data to achieve scientific objectives. Existing analysis methods can quantify parameters such as peak intensity loss in restored images under ideal assumptions. However, determining the shape of the source smearing in dirty images requires using complex observational simulation methods, which limits comprehensive studies of bandwidth smearing from frequency averaging and its optimal application. In this paper, we introduce a semianalytical method (SAM) and validate its accuracy through the simulation method. The SAM efficiently and precisely evaluates the impact of the smearing effect from frequency averaging for various telescope pointing positions, offering a plausible “distorted beam” representation of the non-phase-centered beam shape of point sources in the dirty image. Based on this beam shape, the specific shape of smearing after frequency averaging can be accurately and comprehensively described. This not only enables the determination of the extent of smearing, prediction of accuracy loss, and formulation of effective observation and data processing strategies prior to data processing, but also facilitates the development of more advanced frequency averaging methods.

High-performance multi-catalog fusion or cross-matching has always been an essential issue in astronomical data processing. In this study, we focus on the fusion of multi-band catalog data in a wide-area survey for the China Space Station Telescope (CSST). We implemented a simple and efficient data fusion method based on column-oriented database technology to produce amore consistent and accurate catalog, and this method can carry out the fusion of millions of source records in a few dozen seconds. We analyze and discuss several significant issues related to data fusion, such as the spatial partitioning and indexing of the target sky regions, the efficient implementation of fusion based on joining in the database, and the segmented processing method to address the issue of missing sources at different declinations. The performance profiling results show that by employing the MergeTree table engine within ClickHouse, establishing high-speed indexes based on the spatial partition index number, adopting an appropriate partitioning strategy, and maintaining orderly storage of records in the database according to the spatial partition index number, the efficient fusion of astronomical catalogs can be accomplished through SQL statements. Performance tests show that the proposed method can fulfill CSST data processing requirements, and it is also of reference value for future work related to massive astronomical data fusion. Compared with data fusion systems such as Large Survey DataBase (LSDB), our method can achieve similar performance results with consistent results.

Solar flares, intense solar eruptions, discharge electromagnetic radiation and energetic particles that may have major consequences for both space weather and Earth’s atmospheric conditions. Therefore, developing high-precision forecasting models is crucial. In this paper, we propose a solar flare prediction model, which integrates the Swin Transformer with a TCN augmented by a global attention mechanism, named SwinTCN-Att, for predicting whether ≥C- and ≥M-class flare events will erupt in the solar active regions (ARs) in the next 24 hours. We collected magnetogram data from solar ARs obtained from the Space Weather Helioseismic and Magnetic Imager Active Region Patch (SHARP) dataset, spanning from May 2010 to December 2019, and selected 16 magnetic field feature parameters from the SHARP data. The construction of the model is carried out in two stages: first, the spatial characteristics of the magnetogram are captured using the Swin Transformer; next, these spatial features are integrated with 16 magnetic field parameters. Temporal features are then derived using TCN with a global attention mechanism to predict solar flares. Then, following model training and testing, we evaluated performance using five different assessment metrics, with the True Skill Statistic (TSS) serving as the primary evaluation metric. The results show that the TSS scores achieved were 0.825 ± 0.042 for ≥C-class flares and 0.879 ± 0.025 for ≥M-class flares, marking a significant improvement over previous models. These results demonstrate that the proposed SwinTCN-Att model effectively integrates relevant solar flare information, combines the strengths of both individual models, and captures solar flare evolution features, achieving superior predictive performance.

This research introduces a novel method for fusing multi-view skeleton data to address the limitations encountered by a single vision sensor in capturing motion data, such as skeletal jitter, self-pose occlusion, and the reduced accuracy of three-dimensional coordinate data for human skeletal joints due to environmental object occlusion. Our approach employs two Kinect vision sensors concurrently to capture motion data from distinct viewpoints extract skeletal data and subsequently harmonize the two sets of skeleton data into a unified world coordinate system through coordinate conversion. To optimize the fusion process, we assess the contribution of each joint based on human posture orientation and data smoothness, enabling us to fine-tune the weight ratio during data fusion and ultimately produce a dependable representation of human posture. We validate our methodology using the FMS public dataset for data fusion and model training. Experimental findings demonstrate a substantial enhancement in the smoothness of the skeleton data, leading to enhanced data accuracy and an effective improvement in human posture recognition following the application of this data fusion method.

In multi-touch interfaces, scrolling is a common interactive task. Scrolling performance can help users search and browse content initially off-screen. To evaluate this interaction task by performing analyses and making model assumptions about the interaction process, we divided it into two phases: the search phase and the pointing phase. As a result, quantitative modelling was developed, and two scrolling modes with and without distance feedback were considered. The results of the controlled experiments verified our four proposed mathematical hypotheses of the interaction process and indicated that our model achieved a good fit with and without distance feedback. Our work provides a theoretical foundation for modelling sophisticated scrolling actions and suggestions for user interface designs related to scrolling performance.

The exponential growth of astronomical datasets provides an unprecedented opportunity for humans to gain insight into the Universe. However, effectively analyzing this vast amount of data poses a significant challenge. In response, astronomers are turning to deep learning techniques, but these methods are limited by their specific training sets, leading to considerable duplicate workloads. To overcome this issue, we built a framework for the general analysis of galaxy images based on a large vision model (LVM) plus downstream tasks (DST), including galaxy morphological classification, image restoration, object detection, parameter extraction, and more. Considering the low signal-to-noise ratios of galaxy images and the imbalanced distribution of galaxy categories, we designed our LVM to incorporate a Human-in-the-loop (HITL) module, which leverages human knowledge to enhance the reliability and interpretability of processing galaxy images interactively. The proposed framework exhibits notable few-shot learning capabilities and versatile adaptability for all the abovementioned tasks on galaxy images in the DESI Legacy Imaging Surveys. In particular, for the object detection task, which was trained using 1000 data points, our DST in the LVM achieved an accuracy of 96.7%, while ResNet50 plus Mask R-CNN reached an accuracy of 93.1%. For morphological classification, to obtain an area under the curve (AUC) of similar to 0.9, LVM plus DST and HITL only requested 1/50 of the training sets that ResNet18 requested. In addition, multimodal data can be integrated, which creates possibilities for conducting joint analyses with datasets spanning diverse domains in the era of multi-messenger astronomy.

Radio observation is a method for conducting astronomical observations using radio waves. A common challenge in radio observations is Radio Frequency Interference (RFI), which refers to the unintentional or intentional interference of radio signals from other wireless sources within the radio frequency band. Such interference contaminates the astronomical signals received by radio telescopes, significantly affecting time–frequency domain astronomical observations and research. Consequently, identifying RFI is crucial. In this paper, we employ a deep learning approach to detect RFI present in observation data and propose an improved network structure based on TransUNet. This network leverages the principles of a multi-scale convolutional attention mechanism. It introduces an auxiliary branch to extract high-dimensional image information and an enhanced coordinate attention mechanism for feature map extraction, enabling more comprehensive and accurate identification of RFI in time–frequency images. We introduce a novel architecture named the Multi-Scale TransUNet Network, abbreviated as MS-TransUNet. We utilized observation data from the 40 m radio telescope at the Yunnan Observatory as a data set for training, validating, and testing the network. Compared with previous deep learning networks (U-Net, RFI-Net, R-Net, DSC, EMSCA-UNet), the recall rate and f2 score have been significantly improved. Specifically, the recall rate is improved by at least 2.99%, and the f2 score is improved by at least 2.46%. Experiments demonstrate that this network is exceptional in identifying RFI more comprehensively while ensuring high precision.

Studying the universe through radio telescope observation is crucial. However, radio telescopes capture not only signals from the universe but also various interfering signals, known as Radio Frequency Interference (RFI). The presence of RFI can significantly impact data analysis. Ensuring the accuracy, reliability, and scientific integrity of research findings by detecting and mitigating or eliminating RFI in observational data, presents a persistent challenge in radio astronomy. In this study, we proposed a novel deep learning model called EMSCA-UNet for RFI detection. The model employs multi-scale convolutional operations to extract RFI features of various scale sizes. Additionally, an attention mechanism is utilized to assign different weights to the extracted RFI feature maps, enabling the model to focus on vital features for RFI detection. We evaluated the performance of the model using real data observed from the 40-meter radio telescope at Yunnan Observatory. Furthermore, we compared our results to other models, including U-Net, RFI-Net, and R-Net, using four commonly employed evaluation metrics: precision, recall, F1 score, and IoU. The results demonstrate that our model outperforms the other models on all evaluation metrics, achieving an average improvement of approximately 5\% compared to U-Net. Our model not only enhances the accuracy and comprehensiveness of RFI detection but also provides more detailed edge detection while minimizing the loss of useful signals.

Galaxy mergers exert a pivotal influence on the evolutionary trajectory of galaxies and the expansive development of cosmic structures. The primary challenge encountered in machine learning-based identification of merging galaxies arises from the scarcity of meticulously labeled data sets specifically dedicated to merging galaxies. In this paper, we propose a novel framework utilizing few-shot learning techniques to identify galaxy mergers in the Legacy Surveys. Few-shot learning enables effective classification of merging galaxies even when confronted with limited labeled training samples. We employ a deep convolutional neural network architecture trained on data sets sampled from Galaxy Zoo Decals to learn essential features and generalize to new instances. Our experimental results demonstrate the efficacy of our approach, achieving high accuracy and precision in identifying galaxy mergers with few labeled training samples. Furthermore, we investigate the impact of various factors, such as the number of training samples and network architectures, on the performance of the few-shot learning model. The proposed methodology offers a promising avenue for automating the identification of galaxy mergers in large-scale surveys, facilitating the comprehensive study of galaxy evolution and structure formation. In pursuit of identifying galaxy mergers, our methodology is applied to analyze the Data Release 9 of the Dark Energy Spectroscopic Instrument Legacy Imaging Surveys. As a result, we have unveiled an extensive catalog encompassing 648,183 galaxy merger candidates. We publicly release the catalog alongside this paper.

The ongoing and forthcoming surveys will result in an unprecedented increase in the number of observed galaxies. As a result, data-driven techniques are now the primary methods for analyzing and interpreting this vast amount of information. While deep learning using computer vision has been the most effective for galaxy morphology recognition, there are still challenges in efficiently representing spatial and multi-scale geometric features in practical survey images. In this paper, we incorporate layer attention and deformable convolution into a convolutional neural network (CNN) to bolster its spatial feature and geometric transformation modeling capabilities. Our method was trained and tested on seven classifications of a data set from Galaxy Zoo DECaLS, achieving a classification accuracy of 94.5%, precision of 94.4%, recall of 94.2%, and an F1 score of 94.3% using macroscopic averaging. Our model outperforms traditional CNNs, offering slightly better results while substantially reducing the number of parameters and training time. We applied our method to Data Release 9 of the Legacy Surveys and present a galaxy morphological classification catalog including approximately 71 million galaxies and the probability of each galaxy to be categorized as Round, In-between, Cigar-shaped, Edge-on, Spiral, Irregular, and Error. The code detailing our proposed model and the catalog are publicly available in doi: 10.5281/zenodo.10018255 and GitHub ( https://github.com/kustcn/legacy_galaxy ).

With the development of educational technology, machine learning and deep learning provide technical support for traditional classroom observation assessment. However, in real classroom scenarios, the technique faces challenges such as lack of clarity of raw images, complexity of datasets, multi-target detection errors, and complexity of character interactions. Based on the above problems, a student classroom behavior recognition network incorporating super-resolution and target detection is proposed. To cope with the problem of unclear original images in the classroom scenario, SRGAN (Super Resolution Generative Adversarial Network for Images) is used to improve the image resolution and thus the recognition accuracy. To address the dataset complexity and multi-targeting problems, feature extraction is optimized, and multi-scale feature recognition is enhanced by introducing AKConv and LASK attention mechanisms into the Backbone module of the YOLOv8s algorithm. To improve the character interaction complexity problem, the CBAM attention mechanism is integrated to enhance the recognition of important feature channels and spatial regions. Experiments show that it can detect six behaviors of students—raising their hands, reading, writing, playing on their cell phones, looking down, and leaning on the table—in high-definition images. And the accuracy and robustness of this network is verified. Compared with small-object detection algorithms such as Faster R-CNN, YOLOv5, and YOLOv8s, this network demonstrates good detection performance on low-resolution small objects, complex datasets with numerous targets, occlusion, and overlapping students.

Cancer is one of the most deadly diseases in the world. Accurate cancer subtype classification is critical for patient diagnosis, treatment, and prognosis. Ever-increasing multi-omics data describes the characteristics of the patients from different views and serves as complementary information to promote cancer subtype identification. However, omics data generally have different distributions and high dimensions. How to effectively integrate multiple omics data to classify cancer subtypes accurately is a challenge for researchers. This work proposes a method integrating multi-omics data based on supervised graph contrast learning (MCRGCN) to classify cancer subtypes. The method considers the unique feature distribution of each omics data and the interaction of different omics data features to improve the accuracy of cancer subtype classification. To achieve this, MCRGCN first constructs different sample networks based on the multi-omics data of the samples. Then, it puts the omics data and adjacency matrix of the sample into different residual graph convolution models to get multi-omics features of the samples, which are trained with a supervised comparison loss to maintain that the sample features of each omics should be as consistent as possible. Finally, we input the sample features combining multi-omics features into a classifier to obtain the cancer subtypes. We applied MCRGCN to the invasive breast carcinoma (BRCA) and glioblastoma multiforme (GBM) datasets, integrating gene expression, miRNA expression, and DNA methylation data. The results demonstrate that our model is superior to other methods in integrating multi-omics data. Moreover, the results of survival analysis experiments demonstrate that the cancer subtypes identified by our model have significant clinical features. Furthermore, our model can help to identify potential biomarkers and pathways associated with cancer subtypes.

Open clusters (OCs) are regarded as tracers to understand stellar evolution theory and validate stellar models. In this study, we presented a robust approach to identifying OCs. A hybrid method of pyUPMASK and RF is first used to remove field stars and determine more reliable members. An identification model based on the RF algorithm built based on 3714 OC samples from Gaia DR2 and EDR3 is then applied to identify OC candidates. The OC candidates are obtained after isochrone fitting, the advanced stellar population synthesis (ASPS) model fitting, and visual inspection. Using the proposed approach, we revisited 868 candidates and preliminarily clustered them by the friends-of-friends algorithm in Gaia EDR3. Excluding the open clusters that have already been reported, we focused on the remaining 300 unknown candidates. From high to low fitting quality, these unrevealed candidates were further classified into Class A (59), Class B (21), and Class C (220), respectively. As a result, 46 new reliable open cluster candidates among classes A and B are identified after visual inspection.