Search Results (2,221)

Accurate estimation of the state of health (SOH) and prediction of the remaining useful life (RUL) of lithium-ion batteries (LIBs) are critical for ensuring system reliability and safety across diverse energy storage applications. This paper proposes a hybrid deep learning framework that integrates an axial-embedding Transformer (AxEmbTrans) encoder and a bidirectional LSTM (BiLSTM) module for the joint estimation of SOH and RUL. The AxEmbTrans encoder employs axial attention with abstract embeddings to capture global dependencies among multidimensional health features at reduced computational complexity compared to standard self-attention, while the BiLSTM models local temporal dynamics and short-term degradation fluctuations across consecutive cycles, with its bidirectional structure enhancing robustness against transient noise. Informative health features are extracted from charge–discharge curves, grouped into temporal, energy, and thermal categories, and fused using local linear embedding (LLE) for nonlinear dimensionality reduction. An improved Newton–Raphson-based optimizer (INRBO) is introduced to automatically tune the framework’s key hyperparameters, including the hidden dimension, number of attention heads, number of BiLSTM units, and learning rate, incorporating directional similarity modulation and multi-elite guidance to overcome the convergence instability of the standard NRBO. Extensive experiments on NASA and Maryland datasets demonstrate that the proposed method consistently outperforms baselines in both SOH and RUL prediction, achieving higher accuracy, improved robustness, and better cross-condition generalization. Full article

In the field of industrial robotic vision, accurate recognition and localization of transparent objects pose significant challenges. Unlike opaque objects, transparent objects are difficult to distinguish in RGB images, and due to refraction and reflection, their depth information often suffers from large-area missing or erroneous values, leading to failed grasp pose prediction. Therefore, depth completion is crucial for transparent object grasping tasks. However, existing depth completion methods still exhibit obvious limitations. Multi-stage optimization methods, while achieving high accuracy, involve complex pipelines and high computational costs. Single-stage end-to-end networks, when processing sparse edge features of transparent objects that are also contaminated by background interference, are constrained by the receptive field and smoothing effect of conventional convolutions, often resulting in contour blurring and loss of geometric details. Moreover, existing methods still lack sufficient capability in modeling multi-directional gradient variations of transparent objects under complex backgrounds. To address these issues, this paper proposes ADDFNet for transparent object depth completion, achieving synergistic improvement in accuracy and robustness through two key designs: MDAM and CMFR. To tackle the problem of sparse edge features of transparent objects that are easily disturbed by noise, we design the Multi-directional Differential Attention Module (MDAM), which explicitly extracts multi-directional gradient information through multi-branch differential convolution. Within MDAM, we introduce the Detail Enhancement Differential sub-Module (DEDM) and the Dynamic Convolution with Symmetry-enhanced Geometry Attention sub-module (DSCA) to enhance the network’s perception of fine contours and improve global–local synergistic modeling capability. To address insufficient cross-modal information interaction, we introduce the Cross-Modal Feature Refinement (CMFR) module, which utilizes RGB context to guide and enhance depth features layer by layer during the encoding stage, improving the accuracy and robustness of depth completion while mitigating feature degradation caused by traditional simple fusion approaches. Experimental results on the ClearPose and TransCG datasets demonstrate that ADDFNet outperforms comparison methods in terms of RMSE, REL, MAE, and threshold accuracy metrics, exhibiting more stable performance in edge recovery and internal detail reconstruction of transparent objects. Full article

To address the limitations of traditional methods in feature extraction and multi-modal information fusion, this paper proposes an infrared–visible image object detection architecture that integrates Convolutional Neural Networks (CNNs) and Deformable Transformers. This method leverages the advantages of CNN in local feature modeling and the capabilities of Transformer in capturing global contextual information, facilitating the fusion of semantic consistency and structural details across modalities. By introducing a detection-aware multi-task optimization mechanism, the model improves object detection in challenging scenarios such as low-light conditions, occlusion, and complex backgrounds. Experiments on multiple standard datasets, including M3FD and LLVIP, indicate that the proposed method achieves competitive or better performance than the compared methods in key metrics such as mAP. Specifically, our method obtains the best mAP50 among the evaluated methods with an mAP50 of 74.2% on the M3FD dataset and 98.6% on the LLVIP dataset, surpassing the second-best PIAFusion by 4.3% and 2.5% respectively. These quantitative results support the practicality and effectiveness of our approach in the evaluated complex environments. Full article

To address the problem of road crack detection in low-quality pavement images, existing semantic segmentation methods still have shortcomings such as missed crack detection and inaccurate localization due to weak crack boundaries, low contrast, and complex pavement texture. To address these limitations, this study proposes a crack segmentation model based on feature integration and a triple attention mechanism. The model uses DeepLabv3+ as the backbone network and introduces the proposed three-dimensional interactive attention module after feature extraction. The attention module enhances the extraction of key features related to the spatial location and morphological details of cracks, thereby improving the ability of crack location. A hierarchical feature integration branch is introduced in the cross-layer connection, and a dimension-aware selective fusion module is used to enhance the saliency of small cracks in complex backgrounds. In addition, the proposed multi-group dilation feature fusion module is introduced to improve the multi-scale modeling of small and slender cracks and reduce background interference. The experimental results on Crack500 and GAPS384 datasets show that the proposed model achieves better overall segmentation performance than the comparison model, especially in reducing the missed detection of weak, small, and discontinuous cracks in low-quality pavement images. Complexity analysis further shows that the proposed model maintains practical inference efficiency rather than relying on too large a model size. These results show that the proposed method provides an effective solution for low-quality road crack segmentation, but it still needs to be further verified in actual detection scenarios. Full article

Background/Objectives: Epiretinal membrane (ERM) is a common vitreoretinal disorder characterized by fibrocellular proliferation on the inner retinal surface, often leading to progressive visual impairment. Accurate grading of ERM severity using optical coherence tomography (OCT) is critical for treatment planning and surgical decision-making; however, manual grading is labor-intensive and subjective. This study aims to develop an automated and reliable deep learning-based method for ERM severity classification. Methods: We propose DenseViT-OCT, a hybrid deep learning model that integrates dense convolutional neural networks (CNN) and vision transformers (ViT). The model introduces three key modules: Multi-Scale Dense Feature Aggregation (MDFA) for capturing hierarchical features across multiple spatial scales, Adaptive Feature Calibration (AFC) for enhancing feature discrimination through channel and spatial attention, and Cross-Attention Feature Fusion (CAFF) for enabling bidirectional interaction between convolutional and transformer representations. The model was trained and evaluated on 2195 OCT B-scan images obtained from 397 patients. Results: DenseViT-OCT achieved an overall accuracy of 94.76% on the internal four-class test set, outperforming 19 benchmark models, including ConvNeXt, EfficientNet, ViT, and Swin Transformers. The model demonstrated balanced performance with a macro-averaged precision of 93.76%, recall of 93.22%, F1-score of 93.47%, Cohen’s kappa of 92.62%, and macro-Area Under the Curve (AUC) of 98.95%. Ablation experiments confirmed the contribution of the proposed MDFA, AFC, CAFF, and deep supervision components, with the full model consistently outperforming reduced variants and standalone DenseNet121 and ViT-B/16 backbones. In repeated experiments across five random seeds, DenseViT-OCT also achieved the best mean accuracy (0.9399 ± 0.0052). External validation on the public multicenter OCTDL dataset, performed as binary ERM-versus-normal classification because of label availability, yielded 90.76% accuracy and 97.61% AUC, indicating promising generalization beyond the development cohort. Conclusions: DenseViT-OCT provides a robust framework for automated ERM severity classification from OCT B-scans. The combination of local CNN features, global transformer context, and dedicated fusion modules improves classification performance and yields clinically meaningful error patterns. Although further stage-wise multicenter validation, volumetric OCT analysis, and prospective clinical assessment are required, the proposed method shows promise as a research-oriented decision-support framework for B-scan-level ERM assessment. Full article

Accurate tomato maturity detection is critical for optimizing key agricultural operations in precision agriculture, including harvesting, grading, and quality control. Despite advances in deep learning and machine vision, reliable detection in real-world environments remains challenging due to cluttered backgrounds, dense fruit clustering, and subtle color differences between maturity stages. In response to these challenges, we present TAA-YOLOv8, an attention-enhanced detection architecture integrating a novel Tomato-Adaptive Attention (TAA) module that performs sequential channel–spatial feature refinement using an adaptive 1D convolution for channel recalibration and a balanced 5 × 5 spatial kernel for improved localization, enhancing discriminative representation while preserving computational efficiency. The framework is evaluated on three datasets representing diverse agricultural environments: a newly introduced Cross-Regional Tomato dataset collected from open-field farms in Bangladesh and greenhouse facilities in Japan, and two public benchmarks, Laboro Tomato and Tomato Plantfactory. TAA-YOLOv8m outperforms baseline YOLOv8m, achieving mAP@50–95 improvements of +9.29%, +9.00%, and +6.65% with F1-scores of 0.968, 0.976, and 0.955, respectively. It further surpasses attention-enhanced variants and RT-DETR-L, and remains competitive with YOLOv11m. Gradient-Weighted Class Activation Mapping (Grad-CAM) shows concentrated fruit-centered activations, providing transparent decision-making evidence and supporting stakeholder confidence in practical deployment within vision-based agricultural management systems. Full article

Iris sanguinea features upright, stiff leaves, making it an excellent cut-foliage material, with its tall leaf architecture greatly enhancing ornamental value in landscaping. However, during the leaf expansion phase, plants frequently exhibit loose foliage arrangement, excessive spreading, and compromised mechanical strength, culminating in lodging and a concomitant decline in ornamental quality. Plant height in I. sanguinea is strongly regulated by phytohormones. This study showed that exogenous GA at concentrations of 50 mg·L⁻¹, 100 mg·L⁻¹, and 200 mg·L⁻¹ increased seedling height by 5.7%, 8.8%, and 12.7%, respectively, through foliar spraying on I. sanguinea seedlings grown ex vitro in a greenhouse; conversely, PAC treatment at equivalent concentrations suppressed growth by 19.3%, 21.0%, and 22.2%, respectively. Two pivotal GA signaling components, GAI and GID1a, were isolated from I. sanguinea. Subcellular localization confirmed that both IsGAI and IsGID1a proteins localize to the nucleus. Overexpression vectors pCAMBIA1300-IsGAI-GFP and pCAMBIA1300-IsGID1a-GFP were constructed and expressed in Arabidopsis thaliana. Transgenic lines overexpressing IsGAI showed significantly reduced plant height, hypocotyl elongation, and bolting, whereas IsGID1a overexpression promoted these traits. Exogenous GA application partially reversed the dwarf phenotype induced by IsGAI overexpression and further potentiated the height enhancement observed in IsGID1a-overexpressing lines. This study identifies two key genes controlling plant height and provides a theoretical basis and genetic resources for precisely engineering plant architecture in I. sanguinea. This is especially important for developing dwarf varieties with enhanced ornamental and agronomic traits, offering significant potential in the landscaping and cut flower industries. Full article

Against the backdrop of complex features and spectral heterogeneity in high-resolution remote sensing imagery, traditional methods suffer from insufficient semantic understanding, while existing vision–language change detection models face low efficiency, poor spatial localization, and decoupled detection–description pipelines. To overcome these limitations, this paper proposes ChangeVLM, a language-guided semantic alignment framework for binary remote sensing change detection, enabling end-to-end, prompt-free, highly efficient, and interpretable change detection. Its key advantages include the following, (1) Higher detection accuracy with F1 scores of 91.52%, 83.56%, and 75.29% on LEVIR-CD, SYSU-ChangeDet, and HRCUS datasets, outperforming 18 state-of-the-art methods. (2) Stronger edge integrity and small-object detection capability; (3) practical deployment efficiency: the end-to-end FLOPs is 560.7G. Additionally, under an optimized inference setting with pre-extracted features, the effective computation can be reduced to 13.05G. (4) Language-guided semantic regularization to enhance visual discrimination, without requiring external text prompts. The Asymmetric Fusion Module (AFM), lightweight ChangeHead, and Change-Aware Cross-Modal Fusion Module (CACMF) jointly enhance spatial precision, efficiency, and interpretability. Extensive experiments validate that ChangeVLM achieves a superior accuracy–efficiency trade-off. This method provides an effective, deployable solution for high-resolution remote sensing binary change detection, where the language branch acts only as a regularization signal. Full article

Aims: Coronary heart disease (CHD) associated with rheumatoid arthritis (RA) is a primary driver of mortality in RA patients. In this study, we sought to establish a combined rat model of CHD and RA by integrating cardiac ischemia/reperfusion (I/R), high-fat diet (HFD), and intradermal administration of bovine type II collagen emulsified in complete Freund’s adjuvant. The aim of constructing this model is to investigate and analyze the pathogenesis of RA-induced CHD under the modulation of HFD and cardiac I/R exposure. Methods and Results: Sixty-four male Sprague–Dawley rats were randomly categorized into eight groups (n = 8 per group): control, I/R, HFD, collagen-induced arthritis (CIA), I/R + CIA, HFD + CIA, I/R + HFD, and I/R + HFD + CIA groups (n = 8 per group). We applied Synchrotron radiation-based X-ray micro-computed tomography (micro-CT) to observe the structural changes within the model over time. To further elucidate molecular mechanisms, transcriptome RNA-seq analysis was carried out to identify key signaling pathways, with particular emphasis on the homeostasis of Toll-like receptor 4 (TLR4)/Myd88 signaling in the ischemic myocardium. Furthermore, we conducted in vivo shRNA-mediated knockdown of polymerase I and transcription release factor (PTRF) and evaluated the co-localization of PTRF and TLR4 through immunofluorescence experiments. It is worth mentioning that our rat model of RA-induced (CHD) under a high-fat diet effectively manifested the relevant pathological features that align with the Traditional Chinese Medicine (TCM) definition of “bi” syndrome. The results indicate that the combined stimulation of HFD and CIA significantly elevated cardiac injury markers (CK-MB, LDH, CRP, and c-TNT) and was accompanied by a more severe expansion of the infarct area and increased cardiomyocyte apoptosis compared to the I/R group alone. In addition, the histopathological evaluation revealed significantly aggravated myocardial inflammation and fibrosis deposition, accompanied by extensive areas of tissue damage, further indicating a state of heightened inflammation and severe cardiac degenerative changes. Consistently, myocardial tissues from rats in the I/R + CIA + HFD group exhibited robust activation of the TLR4/MyD88 signaling pathway and a pronounced elevation in the p-JNK/JNK ratio. Moreover, pronounced co-localization between PTRF and TLR4 was evident in small vessels surrounding the infarcted myocardium. Importantly, AAV-mediated knockdown of PTRF attenuated the HFD- and CIA-induced exacerbation of myocardial injury in I/R rats. Conclusions: We successfully established a rat model of CHD with rheumatic syndrome using I/R in combination with RA and HFD. The present findings suggest that the PTRF-related TLR4/MyD88-JNK signaling pathway may act as an important regulatory mechanism underlying myocardial injury aggravated by combined HFD and CIA stimulation. Full article

Grassland fire is one of the major disasters threatening regional ecological security. Its occurrence, development, and spread are closely related to the spatial distribution and coverage of surface vegetation and snow cover across grassland areas. As the primary combustible fuel source, higher vegetation coverage increases fuel load and continuity, thereby directly determining grassland fire danger levels and accelerating fire spread velocity. In contrast, snow cover imposes an indirect regulatory effect on the spatiotemporal pattern of fire danger factors: it lowers surface temperature, raises near-surface humidity, and restricts the germination and growth of herbaceous vegetation in cold seasons, which effectively reduces available combustible materials and weakens regional fire hazard conditions. Therefore, accurately obtaining the coverage status of vegetation (direct combustible fuel factor) and snow cover (indirect fire-regulating factor) in complex grassland scenarios is the essential premise for reliable grassland fire danger monitoring, early warning, disaster prevention and control, and regional ecological management. Aiming at the practical problems in complex grassland scenarios (such as undulating terrain, uneven vegetation growth, large differences in snow depth, and complex lighting conditions), including difficulty in extracting vegetation and snow-covered areas, blurred and confusing boundaries, and low accuracy in coverage calculation, which seriously restrict the technical bottleneck of precise monitoring of grassland fire danger factors, this study takes near-ground images collected by grassland fire danger factor monitoring stations as the core data source, and proposes an improved UNet image segmentation model combined with image segmentation technology and deep learning methods to realize precise extraction of vegetation and snow-covered areas and efficient calculation of coverage in complex scenarios. To improve the model’s feature extraction ability, boundary localization accuracy, and reduce model parameters and computational overhead, the CBAM-ASPP (Convolutional Block Attention Module—Atrous Spatial Pyramid Pooling) module is integrated at the end of the encoding path. The attention mechanism is used to enhance the weight of key features, and the multi-scale receptive field of atrous spatial pyramid pooling is utilized to strengthen the model’s ability to fuse features of vegetation and snow areas of different scales. The residual attention mechanism is introduced in the upsampling stage to effectively alleviate the gradient disappearance problem, improve the model’s ability to accurately locate the boundaries of vegetation and snow areas, and reduce segmentation errors. In the training process, a dynamically weighted hybrid loss function is adopted to dynamically adjust the weights according to the segmentation difficulty of different types of samples during training, optimize the model training effect, and improve the segmentation accuracy and generalization ability. Experiments were conducted using near-ground images of typical complex grassland scenarios as the dataset, and the performance of the proposed model was verified through comparative experiments. The results show that in the vegetation segmentation task, the mean Intersection over Union (mIoU) of the model reaches 84.70%, and the accuracy rate is 91.28%, which are 1.48 and 1.58 percentage points higher than those of the standard UNet model, respectively. In the snow segmentation task, the mIoU of the model reaches 92.74%, and the accuracy rate is 94.19%, which are 2.39 and 2.36 percentage points higher than those of the standard UNet model, respectively. At the same time, the number of parameters of the model is reduced by 12.85% compared with the standard UNet. Also, its comprehensive performance is significantly better than that of mainstream image segmentation models such as FCN, SegNet, and DeepLabv3+. Based on the standardized time-series data retrieved by the optimized segmentation model, this study further constructs a Grassland Fire Risk Index (GFRI) using the Analytic Hierarchy Process (AHP). Pearson correlation verification confirms that the GFRI has an extremely significant positive correlation with historical fire frequency, accurately capturing the seasonal dynamic rhythm of regional grassland fire occurrence. This integrated framework of intelligent segmentation and fire risk quantification provides a reliable technical solution for grassland fire factor monitoring, dynamic fire risk assessment, early warning systems, and refined regional ecological management. Full article

This study investigates the porous structure of Royal Water Lily Leaf vein cross-sections, integrating macroscopic structural observations, quasi-static compression experiments, and finite element simulations to systematically explore the influence of gradient fractal characteristics on mechanical performance and energy absorption behavior. First, the geometric features of the vein cross-sections were extracted through macroscopic measurements, and a parametric model incorporating key variables-porosity, pore ellipticity, and distribution density coefficient-was established. Single-factor analysis reveals that porosity plays a dominant role in determining the overall load-bearing capacity and energy absorption capability; pore ellipticity primarily affects local deformation modes and plateau-stage stability; while the distribution density coefficient significantly regulates the progressive and uniform deformation behavior. Subsequently, a multi-factor coupling model based on the Box–Behnken response surface methodology was developed to investigate the interactions among structural parameters. The results showed that the three variables exhibited significant synergistic effects rather than simple monotonic relationships. Within the investigated range, the optimized configuration (porosity = 30%, ellipticity = 1.6, distribution density coefficient = 1.5) achieved excellent comprehensive performance, with SEA = 115.75 J/kg, MCF = 248.2 N, and CFE = 0.445. Further analysis revealed that the porous vein structure does not exhibit strict self-similar fractal geometry but instead presents a gradient fractal characteristic with hierarchical progression and regional heterogeneity. During compression, the structure undergoes progressive collapse from the inner region outward, enabling staged load-bearing and efficient energy dissipation. These findings provide theoretical support and engineering guidance for the design and optimization of lightweight bioinspired porous energy-absorbing structures. Full article

Intelligent mechanical fault diagnosis plays a key role in maintaining rotating machinery. Although data-driven unsupervised domain adaptation methods have achieved considerable progress, their industrial applications are often restricted by low-quality sensor data. Non-stationary vibration signals and background noise easily corrupt target pseudo-labels, while conventional methods focusing on global statistical matching usually neglect local structures, leading to confirmation bias under dynamic loads. To improve diagnostic reliability, we propose a Noise-Resilient Whitened Domain Adaptation (NRWDA) framework. To handle covariance fluctuations caused by changing working conditions, a Lipschitz-bounded Temporal Whitening (LTW) module is designed as a low-pass filter. An Entropy-guided Prototype Truncation (EPT) mechanism is adopted to discard ambiguous labels and better calibrate semantic centers. In addition, a Dispersion-Adaptive Contrastive Sharpening (DACS) strategy is introduced to dynamically adjust the contrastive temperature based on predictive dispersion, thus tightening decision boundaries. The proposed method is evaluated on CWRU, PU, and MFPT datasets. The PU dataset, featuring fluctuating loads and non-stationary signals, poses a strict test, yet our model maintains its stability even at a 0 dB SNR—a condition where standard approaches usually break down. During the $P0\to P3$ transfer task involving substantial radial force variations, NRWDA secures a 72.36% accuracy and surpasses established baselines. These findings confirm that our technique successfully isolates dependable diagnostic features from corrupted sensor measurements within actual industrial settings. Full article

Point cloud scene flow estimation aims to capture the three-dimensional motion of each point in a sequence of point clouds. Although progress has occurred in this field, existing methods often face significant challenges. In particular, two key issues persist: the absence of corresponding local information from the source point cloud to the target, preventing correct feature matching, and the presence of highly similar adjacent structures in target regions, which leads to ambiguous correspondences due to indistinguishable point features. To address these problems, this paper introduces a novel self-supervised method for point cloud scene flow estimation. Theoretically, we establish a new framework that integrates discriminative feature learning with probabilistic flow refinement. A new network architecture, SetConv++, is designed to learn more discriminative point feature representations, enhancing differentiation in similar structures. Additionally, a refinement module uses the random walk algorithm to adjust initial flow estimates. This approach reconstructs low-confidence flows with high-confidence surrounding ones, reducing missing correspondence issues. Crucially, a new flow smoothing loss term ensures local consistency while suppressing error propagation—a fundamental limitation in existing methods. Through comprehensive experiments on the KITTI Scene Flow dataset, our method demonstrates superior performance. It significantly outperforms existing self-supervised approaches across multiple standard evaluation metrics. Specifically, on the KITTI Scene Flow dataset, our method reduces the Endpoint Error (EPE) by 13.6% (from 0.0411 to 0.0355) and improves Accuracy Strict (AS) by 2.43 percentage points (from 92.68% to 95.11%) compared to baseline self-supervised approaches, while also reducing the outlier rate (Out) by 1.5 percentage points. This advancement not only provides a robust theoretical framework for handling ambiguous correspondences but also enables more reliable and efficient downstream applications—such as autonomous driving perception systems requiring real-time motion accuracy in complex scenes. Full article

Sweet cherry cracking severely impairs its commercial value and causes huge economic losses, and the accurate real-time detection of fine cracking defects remains a challenging small-target detection task. Traditional manual sorting and conventional machine vision methods suffer from low efficiency and poor robustness, while existing YOLO-based models have limitations in multi-scale feature fusion, local feature discrimination and spatial information retention for cherry cracking detection, and their effectiveness in natural production environments has not been statistically validated. To address these issues, this study proposes YOLO-CY for cherry cracking defect detection. Three key modules were optimized: the C3k2_AdditiveBlock was designed to enhance multi-scale feature extraction, the C2PSA_CGLU module improved the discriminability of local crack features via refined channel attention, and the Efficient Up-Convolution Block replaced traditional upsampling to reduce spatial information loss. Experiments were conducted on a self-constructed dataset of 3662 cherry images acquired on a real sorting line under natural ambient light. The results showed that YOLO-CY achieved an mAP50 of 94.88% and an mAP50-95 of 64.92%, with precision and recall reaching 93.90% and 90.81%, respectively, significantly outperforming mainstream lightweight YOLO models and two-stage detectors. Ablation experiments verified the synergistic effect of the three improved modules, and the model only had a marginal increase in parameters (2.62 M) and GFLOPs (6.60), maintaining lightweight characteristics. YOLO-CY can accurately detect fine, low-contrast and pedicel-overlapping cracks and is suitable for real-time detection on automated cherry-sorting lines, providing a technical solution for intelligent cherry quality inspection. Full article

In recent years, channel state information (CSI)-based sensing technology has gradually attracted widespread attention as a contactless and low-cost approach for robotic arm motion understanding. Despite continuous progress in CSI-based human sensing, existing methods of robotic motion sensing still face two key challenges when directly applied to robotic motion sensing: (1) CSI perturbations induced by robotic arm motion are weak and locally distributed, making fine-grained feature extraction difficult. (2) Discriminative information in long robotic arm motion sequences is sparsely concentrated in a few key intervals, and its adaptive temporal selection and enhancement remain challenging. To address the above challenges, this paper proposes an efficient multi-stage robotic arm motion recognition method (named MSPoolNet). The proposed method consists of three key modules: an adaptive temporal downsampling module, a temporal gating module, and a Transformer-based feature encoding module. Specifically, the adaptive temporal downsampling module processes the raw CSI signal at the input stage to achieve local pattern extraction. The temporal gating module adaptively reweights temporal features, dynamically highlighting key temporal segments while suppressing irrelevant information. The proposed Transformer-based feature encoding module replaces conventional self-attention with pooling operations, enabling global information interaction and fine-grained feature representation in a computationally efficient manner. Extensive experimental results demonstrate that the proposed method achieves state-of-the-art performance on two representative public datasets, maintaining a compact model size with an accuracy exceeding 99%. Full article