Global Feature Extraction Research Articles

Improved RepVGG ground-based cloud image classification with attention convolution

Abstract. Atmospheric clouds greatly impact Earth's radiation, hydrological cycle, and climate change. Accurate automatic recognition of cloud shape based on a ground-based cloud image is helpful for analyzing solar irradiance, water vapor content, and atmospheric motion and then predicting photovoltaic power, weather trends, and severe weather changes. However, the appearance of clouds is changeable and diverse, and their classification is still challenging. In recent years, convolution neural networks (CNNs) have made great progress in ground-based cloud image classification. However, traditional CNNs poorly associate long-distance clouds, making the extraction of global features of cloud images quite problematic. This study attempts to mitigate this problem by elaborating on a ground-based cloud image classification method based on the improved RepVGG convolution neural network and attention mechanism. Firstly, the proposed method increases the RepVGG residual branch and obtains more local detail features of cloud images through small convolution kernels. Secondly, an improved channel attention module is embedded after the residual branch fusion, effectively extracting the global features of cloud images. Finally, the linear classifier is used to classify the ground cloud images. Finally, the warm-up method is applied to optimize the learning rate in the training stage of the proposed method, making it lightweight in the inference stage and thus avoiding overfitting and accelerating the model's convergence. The proposed method is validated on the multimodal ground-based cloud dataset (MGCD) and the ground-based remote sensing cloud database (GRSCD) containing seven cloud categories, with the respective classification accuracy rate values of 98.15 % and 98.07 % outperforming those of the 10 most advanced methods used as the reference. The results obtained are considered instrumental in ground-based cloud image classification.

Literature Analysis on Three Dimensional Object Recognition using Deep Learning

The subject of 3D object identification (segmentation, detection, and classification) has received much research in the areas of computer vision, graphics, and machine learning. Recently, deep learning algorithms have surpassed traditional methods for 3D segmentation problems because of their success in 2D computer vision. As a result, a number of novel methods have been developed and tested on a range of gold-standard datasets.In order for the pattern recognition system to correctly identify the item, the features must be extracted in a form that is compatible with the chosen identification technique. The location from which the features are collected does not matter. Local feature extraction and global feature extraction are the two halves of the object recognition approach.This paper provides a comprehensive analysis of the most recent advances in deep learning-based 3D object recognition. We review the most popular 3D object recognition models and evaluate their salient features. Keywords:Deep Learning, Computer Vision, Object Recognition, 3D Objects

Dual-Branch Multitask Fusion Network for Offline Chinese Writer Identification

Chinese characters are complex and contain discriminative information, meaning that their writers have the potential to be recognized using less text. In this study, offline Chinese writer identification based on a single character was investigated. To extract comprehensive features to model Chinese characters, explicit and implicit information as well as global and local features are of interest. A dual-branch multitask fusion network is proposed that contains two branches for global and local feature extraction simultaneously, and introduces auxiliary tasks to help the main task. Content recognition, stroke number estimation, and stroke recognition are considered as three auxiliary tasks for explicit information. The main task extracts implicit information of writer identity. The experimental results validated the positive influences of auxiliary tasks on the writer identification task, with the stroke number estimation task being most helpful. In-depth research was conducted to investigate the influencing factors in Chinese writer identification, with respect to character complexity, stroke importance, and character number, which provides a systematic reference for the actual application of neural networks in Chinese writer identification.

Cross-scenario capacity estimation for lithium-ion batteries via knowledge query domain mixing-up network

Introduction: Deep learning has demonstrated exceptional prowess in estimating battery capacity. However, its effectiveness is often compromised by performance degradation under a consequence of varying operational conditions and diverse charging/discharging protocols.Methods: To tackle this issue, we introduce the Knowledge Query Domain Mixing-up Network (KQDMN), a domain adaptation-based solution adept at leveraging both domain-specific and invariant knowledge. This innovation enriches the informational content of domain features by segregating the functions of feature extraction and domain alignment, enhancing the efficacy of KQDMN in utilizing diverse knowledge types. Moreover, to identify time-deteriorating features in battery time series data, we employ convolutional operations. These operations are pivotal in extracting multi-scale features from the battery's characteristic curves. Inspired by the Transformer model, we have developed a set of knowledge queries that integrate these multi-scale features seamlessly, thereby enabling extensive global feature extraction. To ensure the retention of domain-specific information, we have instituted two independent feature extraction pathways. Pursuing domain-invariant knowledge, this study introduces cross-attention as a mechanism to connect two domain spaces, effectively diminishing the disparity between source and target distributions.Results and Discussion: This approach is crucial for accurately estimating capacity in batteries with diverse performance characteristics. The practicality and robustness of the proposed method are validated using the MIT battery aging dataset, yielding highly satisfactory outcomes. The Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Coefficient of Determination (R2) for our capacity estimation process are 0.19%, 0.23%, and 0.997, respectively, highlighting the precision and reliability of our approach.

Res-BiANet: A Hybrid Deep Learning Model for Arrhythmia Detection Based on PPG Signal

Arrhythmias are among the most prevalent cardiac conditions and frequently serve as a direct cause of sudden cardiac death. Hence, the automated detection of arrhythmias holds significant importance for assisting in the diagnosis of heart conditions. Recently, the photoplethysmography (PPG) signal, capable of conveying heartbeat information, has found application in the field of arrhythmia detection research. This work proposes a hybrid deep learning model, Res-BiANet, designed for the detection and classification of multiple types of arrhythmias. The improved ResNet and BiLSTM models are connected in parallel, and spatial and temporal features of PPG signals are extracted using ResNet and BiLSTM, respectively. Subsequent to BiLSTM, a multi-head self-attention mechanism was incorporated to enhance the extraction of global temporal correlation features over long distances. The model classifies five types of arrhythmia rhythms (premature ventricular contractions, premature atrial contractions, ventricular tachycardia, supraventricular tachycardia, and atrial fibrillation) and normal rhythm (sinus rhythm). Based on this foundation, experiments were conducted utilizing publicly accessible datasets, encompassing a total of 46,827 PPG signal fragments from 91 patients with arrhythmias. The experimental results demonstrate that Res-BiANet achieved exceptional classification performance, including an F1 score of 86.88%, overall accuracy of 92.38%, and precision, sensitivity, and specificity of 88.46%, 85.15%, and 98.43%, respectively. The outstanding performance of the Res-BiANet model suggests significant potential in supporting the auxiliary diagnosis of multiple types of arrhythmias.

A Building Extraction Method for High-Resolution Remote Sensing Images with Multiple Attentions and Parallel Encoders Combining Enhanced Spectral Information

Accurately extracting pixel-level buildings from high-resolution remote sensing images is significant for various geographical information applications. Influenced by different natural, cultural, and social development levels, buildings may vary in shape and distribution, making it difficult for the network to maintain a stable segmentation effect of buildings in different areas of the image. In addition, the complex spectra of features in remote sensing images can affect the extracted details of multi-scale buildings in different ways. To this end, this study selects parts of Xi’an City, Shaanxi Province, China, as the study area. A parallel encoded building extraction network (MARS-Net) incorporating multiple attention mechanisms is proposed. MARS-Net builds its parallel encoder through DCNN and transformer to take advantage of their extraction of local and global features. According to the different depth positions of the network, coordinate attention (CA) and convolutional block attention module (CBAM) are introduced to bridge the encoder and decoder to retain richer spatial and semantic information during the encoding process, and adding the dense atrous spatial pyramid pooling (DenseASPP) captures multi-scale contextual information during the upsampling of the layers of the decoder. In addition, a spectral information enhancement module (SIEM) is designed in this study. SIEM further enhances building segmentation by blending and enhancing multi-band building information with relationships between bands. The experimental results show that MARS-Net performs better extraction results and obtains more effective enhancement after adding SIEM. The IoU on the self-built Xi’an and WHU building datasets are 87.53% and 89.62%, respectively, while the respective F1 scores are 93.34% and 94.52%.

Learning long- and short-term dependencies for improving drug-target binding affinity prediction using transformer and edge contraction pooling.

The accurate identification of drug-target affinity (DTA) is crucial for advancements in drug discovery and development. Many deep learning-based approaches have been devised to predict drug-target binding affinity accurately, exhibiting notable improvements in performance. However, the existing prediction methods often fall short of capturing the global features of proteins. In this study, we proposed a novel model called ETransDTA, specifically designed for predicting drug-target binding affinity. ETransDTA combines convolutional layers and transformer, allowing for the simultaneous extraction of both global and local features of target proteins. Additionally, we have integrated a new graph pooling mechanism into the topology adaptive graph convolutional network (TAGCN) to enhance its capacity for learning feature representations of chemical compounds. The proposed ETransDTA model has been evaluated using the Davis and Kinase Inhibitor BioActivity (KIBA) datasets, consistently outperforming other baseline methods. The evaluation results on the KIBA dataset reveal that our model achieves the lowest mean square error (MSE) of 0.125, representing a 0.6% reduction compared to the lowest-performing baseline method. Furthermore, the incorporation of queries, keys and values produced by the stacked convolutional neural network (CNN) enables our model to better integrate the local and global context of protein representation, leading to further improvements in the accuracy of DTA prediction.

Explanatory Object Part Aggregation for Zero-Shot Learning.

Zero-shot learning (ZSL) aims to recognize objects from unseen classes only based on labeled images from seen classes. Most existing ZSL methods focus on optimizing feature spaces or generating visual features of unseen classes, both in conventional ZSL and generalized zero-shot learning (GZSL). However, since the learned feature spaces are suboptimal, there exists many virtual connections where visual features and semantic attributes are not corresponding to each other. To reduce virtual connections, in this paper, we propose to discover comprehensive and fine-grained object parts by building explanatory graphs based on convolutional feature maps, then aggregate object parts to train a part-net to obtain prediction results. Since the aggregated object parts contain comprehensive visual features for activating semantic attributes, the virtual connections can be reduced by a large extent. Since part-net aims to extract local fine-grained visual features, some attributes related to global structures are ignored. To take advantage of both local and global visual features, we design a feature distiller to distill local features into a master-net which aims to extract global features. The experimental results on AWA2, CUB, FLO, and SUN dataset demonstrate that our proposed method obviously outperforms the state-of-the-arts in both conventional ZSL and GZSL tasks.

LPST-Det: Local-Perception-Enhanced Swin Transformer for SAR Ship Detection

Convolutional neural networks (CNNs) and transformers have boosted the rapid growth of object detection in synthetic aperture radar (SAR) images. However, it is still a challenging task because SAR images usually have the characteristics of unclear contour, sidelobe interference, speckle noise, multiple scales, complex inshore background, etc. More effective feature extraction by the backbone and augmentation in the neck will bring a promising performance increment. In response, we make full use of the advantage of CNNs in extracting local features and the advantage of transformers in capturing long-range dependencies to propose a Swin Transformer-based detector for arbitrary-oriented SAR ship detection. Firstly, we incorporate a convolution-based local perception unit (CLPU) into the transformer structure to establish a powerful backbone. The local-perception-enhanced Swin Transformer (LP-Swin) backbone combines the local information perception ability of CNNs and the global feature extraction ability of transformers to enhance representation learning, which can extract object features more effectively and boost the detection performance. Then, we devise a cross-scale bidirectional feature pyramid network (CS-BiFPN) by strengthening the propagation and integration of both location and semantic information. It allows for more effective utilization of the feature extracted by the backbone and mitigates the problem of multi-scale ships. Moreover, we design a one-stage framework integrated with LP-Swin, CS-BiFPN, and the detection head of R3Det for arbitrary-oriented object detection, which can provide more precise locations for inclined objects and introduce less background information. On the SAR Ship Detection Dataset (SSDD), ablation studies are implemented to verify the effectiveness of each component, and competing experiments illustrate that our detector attains 93.31% in mean average precision (mAP), which is a comparable detection performance with other advanced detectors.

A multi-scale collaborative fusion residual neural network-based approach for bearing fault diagnosis

In recent years, deep learning techniques have become popular for diagnosing equipment faults. However, their real industrial application performance is hindered by challenges related to noise and variable load conditions that prevent accurate extraction of valid feature information. To tackle these challenges, this paper proposed a novel approach known as the multi-scale collaborative fusion residual neural network (MCFRNN) for bearing fault diagnosis. To begin with, the methodology introduces a multi-scale systolic denoising module designed to extract features at multiple scales while mitigating the influence of noise. Subsequently, a central fusion module is employed to explore the intrinsic correlation among the multiple channels and effectively fuse their respective features. Additionally, a global sensing module is incorporated to enhance the perceptual field of MCFRNN, thereby facilitating the extraction of global features. Furthermore, online label smoothing and AdamP are applied to alleviate overfitting and improve the diagnostic capability of MCFRNN under small sample. Finally, the effectiveness of MCFRNN is verified with two publicly available datasets under complex operational and limited sample conditions. The experimental results show that the proposed method has more excellent diagnostic performance and adaptivity than the existing popular methods.

Multiclass AI-Generated Deepfake Face Detection Using Patch-Wise Deep Learning Model

In response to the rapid advancements in facial manipulation technologies, particularly facilitated by Generative Adversarial Networks (GANs) and Stable Diffusion-based methods, this paper explores the critical issue of deepfake content creation. The increasing accessibility of these tools necessitates robust detection methods to curb potential misuse. In this context, this paper investigates the potential of Vision Transformers (ViTs) for effective deepfake image detection, leveraging their capacity to extract global features. Objective: The primary goal of this study is to assess the viability of ViTs in detecting multiclass deepfake images compared to traditional Convolutional Neural Network (CNN)-based models. By framing the deepfake problem as a multiclass task, this research introduces a novel approach, considering the challenges posed by Stable Diffusion and StyleGAN2. The objective is to enhance understanding and efficacy in detecting manipulated content within a multiclass context. Novelty: This research distinguishes itself by approaching the deepfake detection problem as a multiclass task, introducing new challenges associated with Stable Diffusion and StyleGAN2. The study pioneers the exploration of ViTs in this domain, emphasizing their potential to extract global features for enhanced detection accuracy. The novelty lies in addressing the evolving landscape of deepfake creation and manipulation. Results and Conclusion: Through extensive experiments, the proposed method exhibits high effectiveness, achieving impressive detection accuracy, precision, and recall, and an F1 rate of 99.90% on a multiclass-prepared dataset. The results underscore the significant potential of ViTs in contributing to a more secure digital landscape by robustly addressing the challenges posed by deepfake content, particularly in the presence of Stable Diffusion and StyleGAN2. The proposed model outperformed when compared with state-of-the-art CNN-based models, i.e., ResNet-50 and VGG-16.

CenterNet-Saccade: Enhancing Sonar Object Detection with Lightweight Global Feature Extraction.

Sonar imaging technology is widely used in the field of marine and underwater monitoring because sound waves can be transmitted in elastic media, such as the atmosphere and seawater, without much interference. In underwater object detection, due to the unique characteristics of the monitored sonar image, and since the target in an image is often accompanied by its own shadow, we can use the relative relationship between the shadow and the target for detection. To make use of shadow-information-aided detection and realize accurate real-time detection in sonar images, we put forward a network based on a lightweight module. By using the attention mechanism with a global receptive field, the network can make the target pay attention to the shadow information in the global environment, and because of its exquisite design, the computational time of the network is greatly reduced. Specifically, we design a ShuffleBlock model adapted to Hourglass to make the backbone network lighter. The concept of CNN dimension reduction is applied to MHSA to make it more efficient while paying attention to global features. Finally, CenterNet's unreasonable distribution method of positive and negative samples is improved. Simulation experiments were carried out using the proposed sonar object detection dataset. The experimental results further verify that our improved model has obvious advantages over many existing conventional deep learning models. Moreover, the real-time monitoring performance of our proposed model is more conducive to the implementation in the field of ocean monitoring.

Enhancing four-axis machining center accuracy through interactive fusion of spatiotemporal graph convolutional networks and an error-controlled digital twin system

Mitigating thermal errors constitutes a crucial method for enhancing the machining accuracy of four-axis machining centers. At the heart of effective thermal error control lie the thermal error control platform and a resilient thermal error prediction model. It is imperative to note that thermal errors exhibit intricate dynamic and nonlinear spatiotemporal dependencies. However, prevailing thermal error prediction models tend to primarily focus on temporal features or employ simplistic spatiotemporal characteristics, resulting in diminished accuracy and robustness. Furthermore, the present thermal error compensation system is plagued by a lack of user-friendliness, stemming from its suboptimal execution efficiency. In response to the aforementioned challenges, an innovative approach: the interactive fusion spatiotemporal graph convolutional network is proposed. This novel model is specifically designed to capture the intricate dynamic spatiotemporal dependencies inherent in thermal errors. The interactive fusion spatiotemporal graph convolutional network model consists of three essential components: a bilinear temporal convolutional network, a multi-layer spatiotemporal module, and a linear module. These components work in harmony to comprehensively extract both global and local spatiotemporal features. Subsequently, a mapping relationship between thermal errors and compensation components is established, laying the foundation for theoretical advancements in thermal error compensation within the realm of four-axis machining centers. A digital twin system framework tailored for error control is devised, which leverages cloud-edge computing to enable dynamic control and real-time monitoring of thermal errors. To assess the effectiveness of this digital twin system framework and the interactive fusion spatiotemporal graph convolutional network model, a series of rigorous experiments were conducted. The oriented to error-controlled digital twin system coupled with the interactive fusion spatiotemporal graph convolutional network model yielded exceptional machining accuracy, resulting in minimal geometric disparities of [−3.0 μm, 3.0 μm] for the central hole diameter D and [−3.5 μm, 4.0 μm] for the hole distance H.

A 3D Point Cloud Classification Method Based on Adaptive Graph Convolution and Global Attention.

In recent years, there has been significant growth in the ubiquity and popularity of three-dimensional (3D) point clouds, with an increasing focus on the classification of 3D point clouds. To extract richer features from point clouds, many researchers have turned their attention to various point set regions and channels within irregular point clouds. However, this approach has limited capability in attending to crucial regions of interest in 3D point clouds and may overlook valuable information from neighboring features during feature aggregation. Therefore, this paper proposes a novel 3D point cloud classification method based on global attention and adaptive graph convolution (Att-AdaptNet). The method consists of two main branches: the first branch computes attention masks for each point, while the second branch employs adaptive graph convolution to extract global features from the point set. It dynamically learns features based on point interactions, generating adaptive kernels to effectively and precisely capture diverse relationships among points from different semantic parts. Experimental results demonstrate that the proposed model achieves 93.8% in overall accuracy and 90.8% in average accuracy on the ModeNet40 dataset.

Local to global purification strategy to realize collaborative camouflaged object detection

The purpose of camouflaged object detection is to detect objects in images that are not easily perceived by human eyes. Aiming at the problems of low recognition performance and unsatisfied texture information extraction in the complex environment in the current camouflaged object detection algorithms, we propose to improve the accuracy by simultaneously detecting a group of images containing the same camouflaged category. Therefore, we put forward a novel method termed local to global purification network (LGPNet) for collaborative camouflaged object detection. Our method comprises two main modules: the Local Detail Mining module (LDM) and the Global Intra-group Feature Extraction module (GIFE). The LDM is designed to exploit diversified detail information via different adaptive kernels and receptive field mechanisms locally, and the GIFE module is invented for feature enhancement and multi-level information aggregation. Specifically, the GIFE first utilizes channel attention and spatial attention mechanisms to enhance high-level semantic information and then aggregates the intra-group characteristics by level. Extensive experiments on CoCOD8K dataset and 4 COD benchmark datasets illustrate the effectiveness and superiority of our method compared to SOTAs.

Enhancing Radar Echo Extrapolation by ConvLSTM2D for Precipitation Nowcasting.

Precipitation nowcasting in real-time is a challenging task that demands accurate and current data from multiple sources. Despite various approaches proposed by researchers to address this challenge, models such as the interaction-based dual attention LSTM (IDA-LSTM) face limitations, particularly in radar echo extrapolation. These limitations include higher computational costs and resource requirements. Moreover, the fixed kernel size across layers in these models restricts their ability to extract global features, focusing more on local representations. To address these issues, this study introduces an enhanced convolutional long short-term 2D (ConvLSTM2D) based architecture for precipitation nowcasting. The proposed approach includes time-distributed layers that enable parallel Conv2D operations on each image input, enabling effective analysis of spatial patterns. Following this, ConvLSTM2D is applied to capture spatiotemporal features, which improves the model's forecasting skills and computational efficacy. The performance evaluation employs a real-world weather dataset benchmarked against established techniques, with metrics including the Heidke skill score (HSS), critical success index (CSI), mean absolute error (MAE), and structural similarity index (SSIM). ConvLSTM2D demonstrates superior performance, achieving an HSS of 0.5493, a CSI of 0.5035, and an SSIM of 0.3847. Notably, a lower MAE of 11.16 further indicates the model's precision in predicting precipitation.

Transformer-Based Feature Compensation Network for Aerial Photography Person and Ground Object Recognition

Visible-infrared person re-identification (VI-ReID) aims at matching pedestrian images with the same identity between different modalities. Existing methods ignore the problems of detailed information loss and the difficulty in capturing global features during the feature extraction process. To solve these issues, we propose a Transformer-based Feature Compensation Network (TFCNet). Firstly, we design a Hierarchical Feature Aggregation (HFA) module, which recursively aggregates the hierarchical features to help the model preserve detailed information. Secondly, we design the Global Feature Compensation (GFC) module, which exploits Transformer’s ability to capture long-range dependencies in sequences to extract global features. Extensive results show that the rank-1/mAP of our method on the SYSU-MM01 and RegDB datasets reaches 60.87%/58.87% and 91.02%/75.06%, respectively, which is better than most existing excellent methods. Meanwhile, to demonstrate our method‘s transferability, we also conduct related experiments on two aerial photography datasets.

CTCNet: A CNN Transformer capsule network for sleep stage classification

In this paper, we propose a novel neural network architecture called CTCNet. First, we adopt a multi-scale convolutional neural network (MSCNN) to extract low and high-frequency features, adaptive channel feature recalibration (ACFR) to enhance the model’s sensitivity to important channel features in the feature maps and reduce dependence on irrelevant or redundant features, a multi-scale dilated convolutional block (MSDCB) to capture characteristics of different types among feature channels. Second, we use Transformer to extract global temporal context features. Third, we employ capsule network to capture spatial location relationships among EEG features and refine these features. Besides, the capsule network module is used as our model’s classifier to classify the final results. It is worth noting that our model better solves the problem that previous researches failed to take into account the simultaneous extraction of local features and global temporal context characteristics of EEG signals, and ignored the spatial location relationships between these features. Eventually, we assess our model on three datasets and it achieves better or comparable performance than most state-of-the-art methods.

A Dual-Branch Fusion Network Based on Reconstructed Transformer for Building Extraction in Remote Sensing Imagery.

Automatic extraction of building contours from high-resolution images is of great significance in the fields of urban planning, demographics, and disaster assessment. Network models based on convolutional neural network (CNN) and transformer technology have been widely used for semantic segmentation of buildings from high resolution remote sensing images (HRSI). However, the fixed geometric structure and the local receptive field of the convolutional kernel are not good at global feature extraction, and the transformer technique with self-attention mechanism introduces computational redundancies and extracts local feature details poorly in the process of modeling the global contextual information. In this paper, a dual-branch fused reconstructive transformer network, DFRTNet, is proposed for efficient and accurate building extraction. In the encoder, the traditional transformer is reconfigured by designing the local and global feature extraction module (LGFE); the branch of global feature extraction (GFE) performs dynamic range attention (DRA) based on the idea of top-k attention for extracting global features; furthermore, the branch of local feature extraction (LFE) is used to obtain fine-grained features. The multilayer perceptron (MLP) is employed to efficiently fuse the local and global features. In the decoder, a simple channel attention module (CAM) is used in the up-sampling part to enhance channel dimension features. Our network achieved the best segmentation accuracy on both the WHU and Massachusetts building datasets when compared to other mainstream and state-of-the-art methods.

STGAFormer: Spatial–temporal Gated Attention Transformer based Graph Neural Network for traffic flow forecasting

Traffic flow prediction is a critical component of Intelligent Transportation Systems (ITS). However, the dynamic temporal variations in traffic flow, especially in potential occurrence of unexpected incidents, pose challenges to the prediction of traffic flow. This paper proposes a Spatial–temporal Gated Attention Transformer (STGAFormer) model based Graph Neural Network(GNN), leveraging the encoder architecture of the transformer. The gated temporal self-attention in the model, a novel module, can improve the model’s ability to make long-term predictions and handle sudden traffic incidents by enhancing the extraction of both local and global temporal features. Additionally, this paper proposes a distance spatial self-attention module to extract spatial features, which employs thresholding to selectively identify crucial features from both nearby and distant regions. In this way, the model’s ability to assimilate critical spatial information is promoted. Moreover, our model incorporates a diverse range of inputs, including traffic flow attributes, periodicity, proximity adjacency matrix, and adaptive adjacency matrix. Experiments from four real datasets demonstrate that STGAFormer achieves state-of-the-art performance, especially the MAE value of the PeMS08 dataset experiment is improved by 3.82%. This method offers valuable insights and robust support for future transportation planning.