Spatiotemporal Fusion Research Articles

Reconstruction of Fine-Spatial-Resolution FY-3D-Based Vegetation Indices to Achieve Farmland-Scale Winter Wheat Yield Estimation via Fusion with Sentinel-2 Data

The spatial resolution (250–1000 m) of the FY-3D MERSI is too coarse for agricultural monitoring at the farmland scale (20–30 m). To achieve the winter wheat yield (WWY) at the farmland scale, based on FY-3D, a method framework is developed in this work. The enhanced deep convolutional spatiotemporal fusion network (EDCSTFN) was used to perform a spatiotemporal fusion on the 10 day interval FY-3D and Sentinel-2 vegetation indices (VIs), which were compared with the enhanced spatial and temporal adaptive reflectance fusion model (ESTARFM). In addition, a BP neural network was built to calculate the farmland-scale WWY based on the fused VIs, and the Aqua MODIS gross primary productivity product was used as ancillary data for WWY estimation. The results reveal that both the EDCSTFN and ESTARFM achieve satisfactory precision in the fusion of the Sentinel-2 and FY-3D VIs; however, when the period of spatiotemporal data fusion is relatively long, the EDCSTFN can achieve greater precision than ESTARFM. Finally, the WWY estimation results based on the fused VIs show remarkable correlations with the WWY data at the county scale and provide abundant spatial distribution details about the WWY, displaying great potential for accurate farmland-scale WWY estimations based on reconstructed fine-spatial-temporal-resolution FY-3D data.

Multi-relation spatiotemporal graph residual network model with multi-level feature attention: A novel approach for landslide displacement prediction

Accurate prediction of landslide displacement is crucial for effective early warning of landslide disasters. While most existing prediction methods focus on time-series forecasting for individual monitoring points, there is limited research on the spatiotemporal characteristics of landslide deformation. This paper proposes a novel Multi-Relation Spatiotemporal Graph Residual Network with Multi-Level Feature Attention (MFA-MRSTGRN) that effectively improves the prediction performance of landslide displacement through spatiotemporal fusion. This model integrates internal seepage factors as data feature enhancements with external triggering factors, allowing for accurate capture of the complex spatiotemporal characteristics of landslide displacement and the construction of a multi-source heterogeneous dataset. The MFA-MRSTGRN model incorporates dynamic graph theory and four key modules: multi-level feature attention, temporal-residual decomposition, spatial multi-relational graph convolution, and spatiotemporal fusion prediction. This comprehensive approach enables the efficient analyses of multi-source heterogeneous datasets, facilitating adaptive exploration of the evolving multi-relational, multi-dimensional spatiotemporal complexities in landslides. When applying this model to predict the displacement of the Liangshuijing landslide, we demonstrate that the MFA-MRSTGRN model surpasses traditional models, such as random forest (RF), long short-term memory (LSTM), and spatial temporal graph convolutional networks (ST-GCN) models in terms of various evaluation metrics including mean absolute error (MAE = 1.27 mm), root mean square error (RMSE = 1.49 mm), mean absolute percentage error (MAPE = 0.026), and R-squared (R2 = 0.88). Furthermore, feature ablation experiments indicate that incorporating internal seepage factors improves the predictive performance of landslide displacement models. This research provides an advanced and reliable method for landslide displacement prediction.

Remaining useful life prediction with spatio-temporal graph transform and weakly supervised adversarial network: An application in power components

Remaining useful life (RUL) prediction is crucial in improving the reliability of insulated gate bipolar transistors (IGBTs) and has attracted widespread attention all over the world. Owing to the heterogeneity and complex internal connectivity of the IGBTs, accurate RUL prediction by traditional deep learning (DL) methods remains challenging. Herein, a novel prognostic framework is proposed for their RUL prediction. In brief, the non-smooth spike voltage of IGBTs is transformed into a graph matrix that contains rich correlation information. Then, spatio-temporal fusion graph network (STFGN) is proposed to realize RUL prediction. It contains spatial-scale and temporal-scale extractors, where the former one based on GraphSage can obtain unique IGBTs data structure information, while the latter based on bidirectional long short-term memory (BiLSTM) captures the time-series degradation features of IGBTs. Moreover, the weakly supervised adversarial learning strategy is considered to realize cross-domain prediction using scarce labeled data, which enhance the positive transfer. Four transfer tasks under variable working conditions are considered to prove the proposed method. The experimental results demonstrate that the average SF and RMSE metrics of the proposed method are 0.429 and 0.108, which is superior to other related methods.

A multi-task spatio-temporal fusion network for offshore wind power ramp events forecasting

With the accelerated development of offshore wind farms, the impact of Wind Power Ramp Events (WPRE) on power systems has become more pronounced. Accurate prediction of WPRE is essential for mitigating their adverse effects on the grid. However, current studies on offshore WPRE are limited, especially regarding their spatio-temporal correlations and variability across large wind farm clusters. To address this, a novel forecasting approach using a multi-task spatio-temporal fusion network is presented. This method improves WPRE prediction by integrating spatial and temporal data and utilizing multitask learning to characterize ramp features. Firstly, an index system for WPRE characterization is developed, including metrics such as power change rate and duration. Based on this, an X-means classification method for WPRE is established. Secondly, a spatio-temporal encoder combining graph convolutional networks and temporal attention mechanisms has been proposed to model the spatiotemporal dependencies of meteorological changes in offshore wind farms. Finally, a multitask learning framework has been introduced to predict WPRE characteristic indices and occurrence probabilities. This approach enhances feature extraction efficiency through the joint feedback of multiple related tasks, thereby improving prediction accuracy. A case study using data from five offshore wind farms in Eastern China validates the efficacy of the proposed method.

Three-dimensional intelligent monitoring and early warning technology for tailings ponds based on spatiotemporal fusion of multisource big data.

To solve the difficult problems of tailings dam instability and environmental pollution, multisource information perception, prediction and early warning technology for tailings dams are investigated. Taking a tailings pond in China as an example, a three-dimensional visualization intelligent management platform based on the spatiotemporal fusion of multisource big data is established to realize intelligent real-time monitoring, prediction and early warning of tailings dams. A monitoring system for air-space-ground integration was developed via high-resolution optical image recording, unmanned aerial vehicles (UAVs), radar, video surveillance and displacement sensors. This approach can realize pollution monitoring and efficient identification of high-risk areas. In addition, a machine learning algorithm is used to mine spatiotemporal data in detail. The Alibaba cloud platform was adopted to develop a data integration framework. Multisensor spatiotemporal data from tailings dams and multisource monitoring data from the environment were integrated. The integrated management of tailings dam environmental monitoring, pollution and stability assessment is realized. A stability prediction model for tailings dams based on multisource data is proposed. The temporal and spatial information of various forms of data is analysed from multiple levels and perspectives. Rapid prediction of the disaster situation and stability of tailings ponds is realized. Moreover, a tailings pollution assessment model based on a neural network is integrated with multisource information to establish a multifactor environmental pollution neural network assessment model and a multilevel warning platform. This work can provide technical support for intelligent monitoring and early warning systems for tailings ponds.

Health status assessment of pump station units based on spatio-temporal fusion and uncertainty information

An effective health status assessment (HSA) for pump station units (PSUs) is crucial for accurately determining their real status and providing technical support for safe operational decisions. Due to the limitations of existing data-driven HSA methods, which primarily focus on the temporal dependencies of monitoring signals and fail to explore the complex interconnections among signals comprehensively. Moreover, when constructing performance degradation indices based on linear differences, these methods do not effectively integrate heterogeneous signals, resulting in an incomplete and inaccurate assessment of the overall system degradation. This paper proposes a real-time comprehensive HSA method for PSUs based on multi-source heterogeneous uncertainty information. Initially, a health benchmark model (HBM) is built using CrossGNN, which possesses cross-scale and cross-variable interaction capabilities, to precisely capture the temporal dependencies and dynamic relationships among variables in monitoring signals. Subsequently, key measurement points that reflect the operational status of the PSUs are identified through correlation analysis to establish multi-source evaluation indices. Then, considering the uncertainty in signal changes, a novel health degradation index (HDI) is developed using Mahalanobis distance (MD) and the Gaussian Cloud Model (GCM) to analyze changes in unit status. Furthermore, a weighting calculation method based on the non-dominated sorting genetic algorithm (NSGA-II) is proposed to establish a real-time comprehensive health index (RCHDI) for a thorough assessment of PSUs status. Finally, the effectiveness of the proposed method is validated through a case study using data from a pump station in the South-to-North Water Diversion Project in China. The results show that, compared to other studies, the proposed method significantly improves the stability and smoothness of the state assessment curve, with increases of 21.5% and 47.1% respectively, providing a new perspective for comprehensively assessing the health status of PSUs.V

HASTF: a hybrid attention spatio-temporal feature fusion network for EEG emotion recognition.

EEG-based emotion recognition has gradually become a new research direction, known as affective Brain-Computer Interface (aBCI), which has huge application potential in human-computer interaction and neuroscience. However, how to extract spatio-temporal fusion features from complex EEG signals and build learning method with high recognition accuracy and strong interpretability is still challenging. In this paper, we propose a hybrid attention spatio-temporal feature fusion network for EEG-based emotion recognition. First, we designed a spatial attention feature extractor capable of merging shallow and deep features to extract spatial information and adaptively select crucial features under different emotional states. Then, the temporal feature extractor based on the multi-head attention mechanism is integrated to perform spatio-temporal feature fusion to achieve emotion recognition. Finally, we visualize the extracted spatial attention features using feature maps, further analyzing key channels corresponding to different emotions and subjects. Our method outperforms the current state-of-the-art methods on two public datasets, SEED and DEAP. The recognition accuracy are 99.12% ± 1.25% (SEED), 98.93% ± 1.45% (DEAP-arousal), and 98.57% ± 2.60% (DEAP-valence). We also conduct ablation experiments, using statistical methods to analyze the impact of each module on the final result. The spatial attention features reveal that emotion-related neural patterns indeed exist, which is consistent with conclusions in the field of neurology. The experimental results show that our method can effectively extract and fuse spatial and temporal information. It has excellent recognition performance, and also possesses strong robustness, performing stably across different datasets and experimental environments for emotion recognition.

A new deep learning-based model for reconstructing high-quality NDVI time-series data in heavily cloudy areas: fusion of Sentinel 1 and 2 data

ABSTRACT Reconstructing high-quality Normalized Difference Vegetation Index time series data is essential for ecological and agricultural applications but remains challenging in heavily cloudy areas. Fusing Sentinel SAR and optical data with deep learning could be helpful but is also challenging for stable models due to unstable SAR-NDVI relationships caused by imaging mechanism differences and environmental complexities. In this study, we developed a new Bidirectional Recurrent Imputation for Optical-SAR fusion (BRIOS) model to reconstruct high-quality Sentinel-2 NDVI time series data. BRIOS designs a two-layer recurrent architecture that integrates the autocorrelation of discrete, cloud-free NDVI observations into the model for a more stable SAR-NDVI relationship. Evaluating BRIOS against three baseline methods (GF-SG spatiotemporal fusion, Harmonic regression interpolation, and MCNN-Seq deep learning) across three full Sentinel-2 tiles in reconstructing 8-day NDVI time series, BRIOS consistently outperformed in scenarios of either random or continuously missing data, as evidenced by lower RMSE values (e.g. 0.075 for BRIOS vs. 0.108 for GF-SG vs. 0.143 for Harmonic regression vs. 0.303 for MCNN-Seq), better Edge index, and high linear correlation coefficients (R values up to 0.97). Further ablation experiments revealed that deep integration of NDVI autocorrelation features and SAR temporal change patterns has improved the stability and generalization of BRIOS. Discussions on the model's scalability across various cloud sizes and training dataset sizes affirm its practicality for broad-scale application in vegetation monitoring under challenging cloudy conditions.

Crop type identification using spatio-temporal fusion of multi-source remote sensing data based on time-weighted dynamic time warping

Crop type identification using spatio-temporal fusion of multi-source remote sensing data based on time-weighted dynamic time warping

STFM: Accurate Spatio-Temporal Fusion Model for Weather Forecasting

Meteorological prediction is crucial for various sectors, including agriculture, navigation, daily life, disaster prevention, and scientific research. However, traditional numerical weather prediction (NWP) models are constrained by their high computational resource requirements, while the accuracy of deep learning models remains suboptimal. In response to these challenges, we propose a novel deep learning-based model, the Spatiotemporal Fusion Model (STFM), designed to enhance the accuracy of meteorological predictions. Our model leverages Fifth-Generation ECMWF Reanalysis (ERA5) data and introduces two key components: a spatiotemporal encoder module and a spatiotemporal fusion module. The spatiotemporal encoder integrates the strengths of convolutional neural networks (CNNs) and recurrent neural networks (RNNs), effectively capturing both spatial and temporal dependencies. Meanwhile, the spatiotemporal fusion module employs a dual attention mechanism, decomposing spatial attention into global static attention and channel dynamic attention. This approach ensures comprehensive extraction of spatial features from meteorological data. The combination of these modules significantly improves prediction performance. Experimental results demonstrate that STFM excels in extracting spatiotemporal features from reanalysis data, yielding predictions that closely align with observed values. In comparative studies, STFM outperformed other models, achieving a 7% improvement in ground and high-altitude temperature predictions, a 5% enhancement in the prediction of the u/v components of 10 m wind speed, and an increase in the accuracy of potential height and relative humidity predictions by 3% and 1%, respectively. This enhanced performance highlights STFM’s potential to advance the accuracy and reliability of meteorological forecasting.

A deep learning approach for deriving wheat phenology from near-surface RGB image series using spatiotemporal fusion

Accurate monitoring of wheat phenological stages is essential for effective crop management and informed agricultural decision-making. Traditional methods often rely on labour-intensive field surveys, which are prone to subjective bias and limited temporal resolution. To address these challenges, this study explores the potential of near-surface cameras combined with an advanced deep-learning approach to derive wheat phenological stages from high-quality, real-time RGB image series. Three deep learning models based on three different spatiotemporal feature fusion methods, namely sequential fusion, synchronous fusion, and parallel fusion, were constructed and evaluated for deriving wheat phenological stages with these near-surface RGB image series. Moreover, the impact of different image resolutions, capture perspectives, and model training strategies on the performance of deep learning models was also investigated. The results indicate that the model using the sequential fusion method is optimal, with an overall accuracy (OA) of 0.935, a mean absolute error (MAE) of 0.069, F1-score (F1) of 0.936, and kappa coefficients (Kappa) of 0.924 in wheat phenological stages. Besides, the enhanced image resolution of 512 × 512 pixels and a suitable image capture perspective, specifically a sensor viewing angle of 40° to 60° vertically, introduce more effective features for phenological stage detection, thereby enhancing the model’s accuracy. Furthermore, concerning the model training, applying a two-step fine-tuning strategy will also enhance the model’s robustness to random variations in perspective. This research introduces an innovative approach for real-time phenological stage detection and provides a solid foundation for precision agriculture. By accurately deriving critical phenological stages, the methodology developed in this study supports the optimization of crop management practices, which may result in improved resource efficiency and sustainability across diverse agricultural settings. The implications of this work extend beyond wheat, offering a scalable solution that can be adapted to monitor other crops, thereby contributing to more efficient and sustainable agricultural systems.

Spatio-temporal fusion with motion masks for the moving small target detection from remote-sensing videos

Moving small target detection is an eye-catching research topic, full of great challenges in object detection field. Its primary difficulties exist in (i) the very small size proportion of the target region in background images, (ii) the quite weak contrast to background and (iii) the perception of slight target motion. Currently-existing detection methods mainly utilize the image features from spatial domain. This kind of features is quite weak to small targets. More new feature kinds need to be concerned. In view of these, besides traditional image clues, motion features are currently attracting more and more attention in small target detection. In order to promote detection performance, this paper proposes a Spatio-Temporal Fusion Network (STFNet) with two parallel feature extraction branches for detecting moving small targets. In this network, one branch is designed to capture the traditional semantic features from spatial domain, and the other is devised to perceive the slight target motion features hidden in temporal domain. Meanwhile, to optimize the extraction of spatio-temporal features, a group of motion masks is specially designed to guide feature extractors to pay more precise attention to the positions where small targets appear. Moreover, we design a new bridging module to enhance the cross-domain and cross-scale fusion of spatio-temporal features. Through extensive comparison and ablation experiments on seven sub-datasets, it is proved that our new STFNet is effective and it is obviously superior to the compared ones in detecting moving small targets from satellite sequence images. It achieves a Precision of 0.93, an mAP50 of 71.10% and an F1 of 0.84, evidently higher than existing state-of-the-art methods. Our codes are available at https://github.com/UESTC-nnLab/STF.

Deep Spatio-Temporal Network for Low-SNR Cryo-EM Movie Frame Enhancement.

Cryo-EM in single particle analysis is known to have low SNR and requires to utilize several frames of the same particle sample to restore one high-quality image for visualizing that particle. However, the low SNR of cryo-EM movie and motion caused by beam striking make the task very challenging. Video enhancement algorithms in computer vision shed new light on tackling such tasks by utilizing deep neural networks. However, they are designed for natural images with high SNR. Meanwhile, the lack of ground truth in cryo-EM movie seems to be one major limiting factor of the progress. Hence, we present a synthetic cryo-EM movie generation pipeline, which can produce realistic diverse cryo-EM movie datasets with low-SNR movie frames and multiple ground truth values. Then we propose a deep spatio-temporal network (DST-Net) for cryo-EM movie frame enhancement trained on our synthetic data. Spatial and temporal features are first extracted from each frame. Spatio-temporal fusion and high-resolution re-constructor are designed to obtain the enhanced output. For evaluation, we train our model on seven synthetic cryo-EM movie datasets and infer on real cryo-EM data. The experimental results show that DST-Net can achieve better enhancement performance both quantitatively and qualitatively compared with others.

Efficient fusion of spatio-temporal saliency for frame wise saliency identification

<p>Video saliency detection is a rapidly growing subject that has seen very few contributions. The most common technique used nowadays is to perform frame-by-frame saliency detection. The modified Spatio-temporal fusion method presented in this paper offers a novel approach to saliency detection and mapping. It uses frame-wise overall motion color saliency as well as pixel-based consistent Spatio-temporal diffusion for its temporal uniformity. Additionally, a variety of techniques is advocated as a way to increase the saliency maps' overall accuracy and precision. The video is divided into groups of frames, and each frame temporarily goes through diffusion and integration in order to compute the color saliency mapping, as covered in the proposed method section. Then, with the aid of a permutation matrix, the inter-group frame is used to format the pixel-based saliency fusion, after which the features, or the fusion of pixel saliency and color information, direct the diffusion of the spatiotemporal saliency. The result is tested using five publicly accessible global saliency evaluation metrics, and it is determined that the proposed algorithm outperforms numerous saliency detection techniques with an improvement in accuracy margin. The robustness, dependability, adaptability, and precision are all demonstrated by the results.</p>

Daily PM2.5 concentration prediction based on variational modal decomposition and deep learning for multi-site temporal and spatial fusion of meteorological factors.

Air pollution, particularly PM2.5, has long been a critical concern for the atmospheric environment. Accurately predicting daily PM2.5 concentrations is crucial for both environmental protection and public health. This study introduces a new hybrid model within the "Decomposition-Prediction-Integration" (DPI) framework, which combines variational modal decomposition (VMD), causal convolutional neural network (CNN), bidirectional long short-term memory (BiLSTM), and attention mechanism (AM), named as VCBA, for spatio-temporal fusion of multi-site data to forecast daily PM2.5 concentrations in a city. The approach involves integrating air quality data from the target site with data from neighboring sites, applying mathematical techniques for dimensionality reduction, decomposing PM2.5 concentration data using VMD, and utilizing Causal CNN and BiLSTM models with an attention mechanism to enhance performance. The final prediction results are obtained through linear aggregation. Experimental results demonstrate that the VCBA model performs exceptionally well in predicting daily PM2.5 concentrations at various stations in Taiyuan City, Shanxi Province, China. Evaluation metrics such as RMSE, MAE, and R2 are reported as 2.556, 1.998, and 0.973, respectively. Compared to traditional methods, this approach offers higher prediction accuracy and stronger spatio-temporal modeling capabilities, providing an effective solution for accurate PM2.5 daily concentration prediction.

An interpretable hybrid spatiotemporal fusion method for ultra-short-term photovoltaic power prediction

An interpretable hybrid spatiotemporal fusion method for ultra-short-term photovoltaic power prediction

Exploring the optimal sequences for fusing land surface temperature via spatiotemporal fusion: ‘retrieval-then-fusion’ or ‘fusion-then-retrieval’?

ABSTRACT Due to the limitations inherent in satellite sensor capabilities, it remains a challenge to balance the temporal and spatial resolutions of current remote sensing observations. Spatiotemporal fusion has emerged as a viable approach to address this issue. Many spatiotemporal fusion models have been proposed to fuse surface reflectance initially; however, their adaption for land surface temperature (LST), particularly with the ‘fusion-then-retrieval (FTR)’ or ‘retrieval-then-fusion (RTF)’ strategy, has not yet been comprehensively explored. This paper describes a comparative study of the two strategies within the context of LST fusion using three classic fusion methods including the spatial and temporal adaptive reflectance fusion model (STARFM), the enhanced spatial and temporal adaptive reflectance fusion model (ESTARFM) and the flexible spatiotemporal data fusion model (FSDAF). The findings illuminate nuanced applicability across different models: (1) the FTR strategy is more suitable for STARFM; (2) the strategic applicability for ESTARFM and FSDAF is strongly related to the reference images; (3) the STARFM and FSDAF models are easily affected by the time interval between the predicted date and the reference date. This work is expected to provide practical guidance for synthesizing high spatiotemporal LST data with fusion models.

A fast adaptive spatio-temporal fusion method to enhanced Fit-FC.

Space-time fusion is an economical and efficient way to solve "space-time contradiction". Among all kinds of space-time fusion methods, Fit-FC space-time fusion method based on weight Function is widely used. However, this method is based on the linear model to depict the phase change, but the phase change in the real scene is complicated, and the linear model is difficult to accurately capture the phase change, resulting in the spectral distortion of the fusion image. In addition, pixel-by-pixel scanning with moving Windows leads to inefficiency issues, limiting its use in large-scale and long-term tasks. To overcome these limitations, this paper developed a simple and fast adaptive remote sensing image Spatio-Temporal fusion method based on Fit-FC, called Adapt Lasso-Fit-FC (AL-FF). Firstly, the sparse characteristics of time phase change between images are explored, and a time phase change estimation model based on sparse regression is constructed, which overcomes the fuzzy problem of fusion image caused by the failure of linear regression to capture complex nonlinear time phase transition in the weighted Function method, making the algorithm better at capturing details. Secondly, an adaptive window selection Function is established to overcome the problem of manually setting parameters on different data sets, improve the convenience of the algorithm and robustness of the application on different data sets, and make the algorithm simpler and more efficient. Finally, the improved AL-FF algorithm is compared with other algorithms to verify the performance improvement. Compared with the current advanced Spatio-Temporal fusion methods, AL-FF algorithm has stronger detail capture ability and can generate more accurate fusion results. In addition, the computational efficiency is significantly improved, and the efficiency is increased by more than 20 times compared with the current mainstream method.

MGSFformer: A Multi-Granularity Spatiotemporal Fusion Transformer for air quality prediction

Air quality spatiotemporal prediction can provide technical support for environmental governance and sustainable city development. As a classic multi-source spatiotemporal data, effective multi-source information fusion is key to achieving accurate air quality predictions. However, due to not fully fusing two pieces of information, classical deep learning models struggle to achieve satisfactory prediction results: (1) Multi-granularity: each air monitoring station collects air quality data at different sampling intervals, which show distinct time series patterns. (2) Spatiotemporal correlation: due to human activities and atmospheric diffusion, there exist correlations between air quality data from different air monitoring stations, necessitating the consideration of other air monitoring stations' influences when modeling each air quality time series. In this study, to achieve satisfactory prediction results, we propose the Multi-Granularity Spatiotemporal Fusion Transformer, comprised of the residual de-redundant block, spatiotemporal attention block, and dynamic fusion block. Specifically, the residual de-redundant block eliminates information redundancy between data with different granularities and prevents the model from being misled by redundant information. The spatiotemporal attention block captures the spatiotemporal correlation of air quality data and facilitates prediction modeling. The dynamic fusion block evaluates the importance of data with different granularities and integrates the prediction results. Experimental results demonstrate that the proposed model surpasses 11 baselines by 5% in performance on three real-world datasets.

Phase-specific augmented reality guidance for microscopic cataract surgery using spatiotemporal fusion network

Phacoemulsification cataract surgery (PCS) is a routine procedure conducted using a surgical microscope, heavily reliant on the skill of the ophthalmologist. While existing PCS guidance systems extract valuable information from surgical microscopic videos to enhance intraoperative proficiency, they suffer from non-phase-specific guidance, leading to redundant visual information. In this study, our major contribution is the development of a novel phase-specific augmented reality (AR) guidance system, which offers tailored AR information corresponding to the recognized surgical phase. Leveraging the inherent quasi-standardized nature of PCS procedures, we propose a two-stage spatiotemporal fusion network for surgical microscopic video recognition. In the first stage, we implement a multi-task learning structure to segment the surgical limbus region and extract limbus region-focused spatial feature for each frame. In the second stage, we propose a spatiotemporal fusion transformer (STFT) network to model local fine-grained and global temporal relationships, and fuse the extracted spatial features to recognize the current surgical phase. Additionally, we collaborate closely with ophthalmologists to design AR visual cues by utilizing techniques such as limbus ellipse fitting and regional restricted normal cross-correlation rotation computation. We evaluated the network on publicly available and in-house datasets, with comparison results demonstrating its superior performance compared to related works. Ablation results further validated the effectiveness of the limbus region-focused spatial feature extractor and the fusion of temporal features. Furthermore, the developed system was evaluated in a clinical setup, with results indicating acceptable accuracy and real-time performance, underscoring its potential for clinical applications.