Search Results (2,470)

The concentrations of carbon, nitrogen, and phosphorus in water bodies significantly influence aquatic ecological conditions. By collecting multitemporal hyperspectral data and water quality parameter data from water bodies and through systematic preprocessing of hyperspectral data combined with multimethod sensitive band selection, an optimal spectral feature subset was determined. Within a machine learning framework, multiple combined remote sensing inversion models were constructed to identify the optimal inversion model for each water quality parameter, along with corresponding preprocessing methods and sensitive bands. The results indicate that differential processing of remote sensing reflectance enhances model accuracy. Sensitive band selection effectively eliminates redundant bands, significantly improving the computational efficiency of inversion models. XGBoost demonstrated superior accuracy in constructing 240 water quality parameter inversion models because of its unique algorithmic design. However, model accuracy is not solely determined by algorithmic complexity or predictive capability but rather by the combined effect of algorithm performance and input feature quality. Verification of the inversion model’s generalization ability via an independent dataset demonstrated its capacity for generalization. These findings provide valuable insights for the reliable application of hyperspectral data in aquatic environmental remote sensing and offer support for regional water quality conservation efforts. Full article

The widespread distribution of digital videos in social networks, streaming services, and surveillance systems has increased the risk of manipulation, unauthorized redistribution, and adversarial tampering. This paper presents a lightweight deep learning framework for robust and imperceptible video watermarking designed specifically for cybersecurity environments. Unlike heavy architectures that rely on multi-scale feature extractors or complex adversarial networks, our model introduces a compact encoder–decoder pipeline optimized for real-time watermark embedding and recovery under adversarial attacks. The proposed system leverages spatial attention and temporal redundancy to ensure robustness against distortions such as compression, additive noise, and adversarial perturbations generated via Fast Gradient Sign Method (FGSM) or recompression attacks from generative models. Experimental simulations using a reduced Kinetics-600 subset demonstrate promising results, achieving an average PSNR of 38.9 dB, SSIM of 0.967, and Bit Error Rate (BER) below 3% even under FGSM attacks. These results suggest that the proposed lightweight framework achieves a favorable trade-off between resilience, imperceptibility, and computational efficiency, making it suitable for deployment in video forensics, authentication, and secure content distribution systems. Full article

Detecting rice leaf diseases in real-world field environments remains challenging due to varying lesion sizes, diverse lesion morphologies, complex backgrounds, and the limited availability of high-quality annotated datasets. Existing detection models often suffer from performance degradation under these conditions, particularly when training data lack sufficient diversity and structural realism. To address these challenges, this paper proposes a Latent Diffusion Inpainting (LDI) data augmentation method and an improved lightweight detection model, LGH-YOLOv12n. Unlike conventional diffusion-based augmentation methods that generate full images or random patches, LDI performs category-aware latent inpainting, synthesizing realistic lesion patterns by jointly conditioning on background context and disease categories, thereby enhancing data diversity while preserving scene consistency. Furthermore, LGH-YOLOv12n improves upon the YOLOv12n baseline by introducing GSConv in the backbone to reduce channel redundancy and enhance lesion localization, and integrating Hierarchical Multi-head Attention (HMHA) into the neck network to better distinguish disease features from complex field backgrounds. Experimental results demonstrate that LGH-YOLOv12n achieves an F1 of 86.1% and an mAP@50 of 88.3%, outperforming the YOLOv12n model trained without data augmentation by 3.3% and 5.0%, respectively. Moreover, when trained on the LDI-augmented dataset, LGH-YOLOv12n consistently outperforms YOLOv8n, YOLOv10n, YOLOv11n, and YOLOv12n, with mAP@50 improvements of 4.6%, 5.2%, 1.9%, and 2.1%, respectively. These results indicate that the proposed LDI augmentation and LGH-YOLOv12n model provide an effective and robust solution for rice leaf disease detection in complex field environments. Full article

Accurate photovoltaic (PV) output power forecasting is essential for reliable power system operation, yet rapidly changing meteorological conditions often degrade forecasting accuracy. This study proposes an attention-enhanced bidirectional gated recurrent unit (BiGRU) optimized by an improved Marine Predators Algorithm (IMPA) for PV output power forecasting. Kernel Principal Component Analysis (KPCA) is first employed to extract compact nonlinear representations and suppress redundant features. Then, a dual multi-head self-attention mechanism is integrated before and after the BiGRU layer to strengthen temporal feature learning under fluctuating weather. Finally, the IMPA is designed to improve exploration–exploitation balance and automatically optimize key hyperparameters. Experiments under sunny, cloudy, and rainy conditions demonstrate that IMPA-Att-BiGRU reduces MAE and RMSE by 35.7–58.5% and 22.8–49.1% versus BiGRU, respectively, while increasing R² by 2.2–4.1 percentage points. Against the best benchmark (LSTM), MAE and RMSE are further reduced by 38.1–49.5% and 33.8–52.4%. Moreover, in a cross-day rolling forecasting test with fivefold results, IMPA-Att-BiGRU achieves 62.4% MAE and 49.3% RMSE reductions over BiGRU, confirming robust performance under long-horizon error accumulation. Full article

Real-time object detection in Unmanned Aerial Vehicle (UAV) imagery presents unique challenges, primarily characterized by extreme scale variations and intense background clutter. Existing detectors often suffer from spectral homogenization in which the critical high-frequency details of minute targets are washed out by dominant background signals during feature downsampling. To address this, we propose Eagle-YOLO, a dynamic feature aggregation framework designed to master these complexities without compromising inference speed. We introduce three core innovations: (1) the Hierarchical Granularity Block (HG-Block), which employs a residual granularity injection pathway to function as a detail anchor for tiny objects while simultaneously accumulating semantics for large structures; (2) the Cross-Stage Context Modulation (CSCM) mechanism, which leverages a global context query to filter background redundancy and recalibrate features across network stages; and (3) the Scale-Adaptive Heterogeneous Convolution (SAHC) strategy, which dynamically aligns receptive fields with the inherent scale distribution of aerial data. Extensive experiments on the DUT Anti-UAV dataset demonstrate that Eagle-YOLO achieves a remarkable balance between accuracy and latency. Specifically, our lightweight Eagle-YOLO-T variant achieves 74.62% AP, surpassing the robust baseline RTMDet-T by 1.67% while maintaining a real-time inference speed of 141 FPS on an NVIDIA RTX 4090 GPU. Furthermore, on the challenging Anti-UAV dataset, our Eagle-YOLOv8-M variant reaches an impressive 94.38% ${\mathrm{AP}}_{50}^{\mathrm{val}}$, outperforming the standard YOLOv8-M by 2.83% and proving its efficacy for edge-deployed aerial surveillance applications. Full article

Heavy precipitation events are characterized by sudden onset, limited spatiotemporal scales, rapid evolution, and high disaster potential, posing long-standing challenges in weather forecasting. With the development of deep learning, an increasing number of researchers have leveraged its powerful feature representation and non-linear modeling capabilities to address the challenge of precipitation nowcasting. Despite recent advances in deep learning for precipitation nowcasting, most existing methods do not explicitly separate precipitation from non-precipitation regions. This often leads to the extraction of redundant or irrelevant features, thereby causing models to learn misleading patterns and ultimately reducing their predictive capability for heavy precipitation events. To address this issue, we propose a novel dynamic region attention (DRA) mechanism, and an improved model TPDTC-Net-DRA, based on our previously introduced TPDTC-Net. The proposed TPDTC-Net-DRA applies the DRA mechanism and incorporates its two key components: a dynamic region module and a weight control module. The dynamic region module generates a mask matrix that is applied to the feature maps, guiding the attention mechanism to focus only on precipitation areas. Meanwhile, the weight control module produces a location-sensitive weight matrix to direct the model’s attention toward regions with intense precipitation. Extensive experiments demonstrate that TPDTC-Net-DRA achieves superior performance for heavy precipitation, outperforming current state-of-the-art methods, and indicate that the proposed DRA mechanism exhibits strong generalization ability across diverse model architectures. Full article

Precise short-term forecasting of wind power generation is indispensable for ensuring the security and economic efficiency of power grid operations. Nevertheless, the inherent non-stationarity and stochastic nature of wind power series present significant challenges for prediction accuracy. To address these issues, this paper proposes a dual-optimized hybrid deep learning framework combining Spearman correlation analysis, RIME-VMD, and TCN-BiGRU-SA. First, Spearman correlation analysis is employed to screen meteorological factors, eliminating redundant features and reducing model complexity. Second, an adaptive Variational Mode Decomposition (VMD) strategy, optimized by the RIME algorithm based on Minimum Envelope Entropy, decomposes the non-stationary wind power series into stable intrinsic mode functions (IMFs). Third, a hybrid predictor integrating Temporal Convolutional Network (TCN), Bidirectional Gated Recurrent Unit (BiGRU), and Self-Attention (SA) mechanisms is constructed to capture both local trends and long-term temporal dependencies. Furthermore, the RIME algorithm is utilized again to optimize the hyperparameters of the deep learning predictor to avoid local optima. The proposed framework is validated using full-year datasets from two distinct wind farms in Xinjiang and Gansu, China. Experimental results demonstrate that the proposed model achieves a Root Mean Square Error (RMSE) of 7.5340 MW on the primary dataset, significantly outperforming mainstream baseline models. The multi-dataset verification confirms the model’s superior prediction accuracy, robustness against seasonal variations, and strong generalization capability. Full article

In recent years, Transformer-based methods have demonstrated proficiency in capturing complex patterns for time series forecasting. However, their quadratic complexity relative to input sequence length poses a significant bottleneck for scalability and real-world deployment. Recently, the Mamba architecture has emerged as a compelling alternative by mitigating the prohibitive computational overhead and latency inherent in Transformers. Nevertheless, a vanilla Mamba backbone often struggles to adequately characterize intricate temporal dynamics, particularly long-term trend shifts and non-stationary behaviors. To bridge the gap between Mamba’s global scanning and local dependency modeling, we propose C-T-Mamba, a hybrid framework that synergistically integrates a Mamba block, channel attention, and a temporal convolution block. Specifically, the Mamba block is leveraged to capture long-range temporal dependencies with linear scaling, the channel attention mechanism filters redundant information, and the temporal convolution block extracts multi-scale local and global features. Extensive experiments on five public benchmarks demonstrate that C-T-Mamba consistently outperforms state-of-the-art (SOTA) baselines (e.g., PatchTST and iTransformer), achieving average reductions of 4.3–18.5% in MSE and 3.9–16.2% in MAE compared to representative Transformer-based and CNN-based models. Inference scaling analysis reveals that C-T-Mamba effectively breaks the computational bottleneck; at a horizon of 1536, it achieves an $8.8\times$ reduction in GPU memory and over 10× speedup compared to standard Transformers. At 2048 steps, its latency remains as low as 8.9 ms, demonstrating superior linear scaling. These results underscore that C-T-Mamba achieves SOTA accuracy while maintaining a minimal computational footprint, making it highly effective for long-term multivariate time series forecasting. Full article

Large volumes of images shared on social media have made image captioning an important tool for social hotspot identification and public opinion monitoring, where accurate visual–language alignment is essential for reliable analysis. However, existing image captioning models based on BLIP-2 (Bootstrapped Language–Image Pre-training) often struggle with complex, context-rich, and socially meaningful images in real-world social media scenarios, mainly due to insufficient cross-modal interaction, redundant visual token representations, and an inadequate ability to capture multi-scale semantic cues. As a result, the generated captions tend to be incomplete or less informative. To address these limitations, this paper proposes ECMA (Enhanced Cross-Modal Attention), a lightweight module integrated into the Querying Transformer (Q-Former) of BLIP-2. ECMA enhances cross-modal interaction through bidirectional attention between visual features and query tokens, enabling more effective information exchange, while a multi-scale visual aggregation strategy is introduced to model semantic representations at different levels of abstraction. In addition, a semantic residual gating mechanism is designed to suppress redundant information while preserving task-relevant features. ECMA can be seamlessly incorporated into BLIP-2 without modifying the original architecture or fine-tuning the vision encoder or the large language model, and is fully compatible with OPT (Open Pre-trained Transformer)-based variants. Experimental results on the COCO (Common Objects in Context) benchmark demonstrate consistent performance improvements, where ECMA improves the CIDEr (Consensus-based Image Description Evaluation) score from 144.6 to 146.8 and the BLEU-4 score from 42.5 to 43.9 on the OPT-6.7B model, corresponding to relative gains of 1.52% and 3.29%, respectively, while also achieving competitive METEOR (Metric for Evaluation of Translation with Explicit Ordering) scores. Further evaluations on social media datasets show that ECMA generates more coherent, context-aware, and socially informative captions, particularly for images involving complex interactions and socially meaningful scenes. Full article

Richter transformation (RT) represents a rare but highly lethal evolution of chronic lymphocytic leukemia/small lymphocytic lymphoma (CLL/SLL), most frequently manifesting as diffuse large B-cell lymphoma (DLBCL). Despite therapeutic advances in CLL, DLBCL-RT remains characterized by rapid progression, profound treatment refractoriness, and short survival with conventional chemoimmunotherapy, underscoring the need for a refined biological and therapeutic framework. A defining feature of RT is clonal relatedness: most cases arise through linear or branched evolution of the antecedent CLL clone and carry an inferior prognosis compared with clonally unrelated cases that resemble de novo DLBCL. Recent multi-omic data further indicate that clonally related RT commonly originates from minute, transformation-primed subclones detectable years before clinical emergence, shifting RT from a late stochastic event to an early-established evolutionary trajectory. At transformation, recurrent genetic lesions of TP53, CDKN2A/B, NOTCH1, and MYC cooperate with B-cell receptor-associated programs, epigenetic reconfiguration, and metabolic rewiring toward OXPHOS- and mTOR-driven states, collectively promoting genomic instability and aggressive growth. In parallel, RT develops within a profoundly immunosuppressive microenvironment marked by PD-1-expressing malignant B cells, PD-L1-rich myeloid niches, exhausted T cells, expanded regulatory T cells, and M2-skewed macrophages interconnected by redundant checkpoint and cytokine networks. Therapeutic strategies are rapidly evolving, including pathway inhibitors, immune checkpoint blockade, T-cell-engaging bispecific antibodies, CAR-T therapies, and antibody–drug conjugates. This review integrates current insights into RT pathogenesis, immune escape, and emerging therapies, highlighting opportunities for biomarker-driven patient stratification, rational combinations, and earlier interception of transformation-prone disease. Full article

Tea bud recognition and localization constitute a fundamental step toward enabling fine-grained tea plantation management and intelligent harvesting, offering substantial value in improving the picking quality of premium tea materials, reducing labor dependency, and accelerating the development of smart tea agriculture. However, most existing methods for detecting tea buds are built upon Artificial Neural Networks (ANNs) and rely extensively on floating-point computation, making them difficult to deploy efficiently on energy-constrained edge platforms. To address this challenge, this paper proposes an energy-efficient tea bud detection model, GAE-SpikeYOLO, which improves upon the Spiking Neural Networks (SNNs) detection framework SpikeYOLO. Firstly, Gated Attention Coding (GAC) is introduced into the input encoding stage to generate spike streams with richer spatiotemporal dynamics, strengthening shallow feature saliency while suppressing redundant background spikes. Secondly, the model incorporates the Temporal-Channel-Spatial Attention (TCSA) module into the neck network to enhance deep semantic attention on tea bud regions and effectively suppress high-level feature responses unrelated to the target. Lastly, the proposed model adopts the EIoU loss function to further improve bounding box regression accuracy. The detection capability of the model is systematically validated on a tea bud object detection dataset collected in natural tea garden environments. Experimental results show that the proposed GAE-SpikeYOLO achieves a Precision (P) of 83.0%, a Recall (R) of 72.1%, a mAP@0.5 of 81.0%, and a mAP@[0.5:0.95] of 60.4%, with an inference energy consumption of only 49.4 mJ. Compared with the original SpikeYOLO, the proposed model improves P, R, mAP@0.5, and mAP@[0.5:0.95] by 1.4%, 1.6%, 2.0%, and 3.3%, respectively, while achieving a relative reduction of 24.3% in inference energy consumption. The results indicate that GAE-SpikeYOLO provides an efficient and readily deployable solution for tea bud detection and other agricultural vision tasks in energy-limited scenarios. Full article

Accurate and non-destructive assessment of plant nutritional status remains a key challenge in precision agriculture, particularly under dynamic physiological conditions such as dehydration. Therefore, this study focused on developing an integrated nutritional assessment framework for avocado (Persea americana Mill.) leaves across progressive dehydration stages using spectral analysis. A novel nutritional function index (NFI) was innovatively constructed using an entropy-weighted multi-criteria decision-making approach. This unified assessment metric integrated critical physiological indicators, such as moisture content, nitrogen content, and chlorophyll content estimated from soil and plant analyzer development (SPAD) readings. To enhance the prediction accuracy and interpretability of NFI, innovative vegetation indices (VIs) specifically tailored to NFI were systematically constructed using exhaustive wavelength-combination screening. Optimal wavelengths identified from short-wave infrared regions (1446, 1455, 1465, 1865, and 1937 nm) were employed to build physiologically meaningful VIs, which were highly sensitive to moisture and biochemical constituents. Feature wavelengths selected via the successive projections algorithm and competitive adaptive reweighted sampling further reduced spectral redundancy and improved modeling efficiency. Both feature-level and algorithm-level data fusion methods effectively combined VIs and selected feature wavelengths, significantly enhancing prediction performance. The stacking algorithm demonstrated robust performance, achieving the highest predictive accuracy (R2V = 0.986, RMSEV = 0.032) for NFI estimation. This fusion-based modeling approach outperformed conventional single-model schemes in terms of accuracy and robustness. Unlike previous studies that focused on isolated spectral predictors, this work introduces an integrative framework combining entropy-weighted feature synthesis and multiscale fusion learning. The developed strategy offers a powerful tool for real-time plant health monitoring and supports precision agricultural decision-making. Full article

Traffic forecasting in large-scale urban networks must operate reliably under imperfect sensing conditions, where measurements may contain noise or missing values. Most existing spatio-temporal graph neural networks focus primarily on modeling spatial–temporal dependencies, while paying limited attention to the propagation of irrelevant or unstable information through dynamic graph structures. In this work, we propose a Dynamic Graph Information Bottleneck (DGIB) framework that enhances prediction stability by introducing task-aware representation compression into dynamic graph learning. Instead of relying solely on architectural complexity, DGIB explicitly regulates the information flow within spatio-temporal embeddings through a variational bottleneck objective. The model adaptively constructs time-evolving adjacency matrices, extracts spatial features via graph convolutions, captures temporal dependencies using recurrent modeling, and constrains the latent representation to retain only predictive content relevant to future traffic states. By jointly optimizing topology adaptation and information-theoretic regularization in an end-to-end manner, the proposed framework mitigates the amplification of noisy or redundant signals in dynamic graphs. Experiments on multiple benchmark traffic datasets demonstrate that DGIB achieves competitive forecasting accuracy while maintaining strong robustness under noisy and incomplete data scenarios. Full article

With the advancement of cotton mechanized topping technology, deep learning-based methods for detecting cotton apical buds have made significant progress in improving detection accuracy. However, existing algorithms generally suffer from complex structures, large parameter counts, and high computational costs, making them difficult to deploy in practical field environments. To address this, this paper proposes a lightweight YOLO-SEW algorithm for detecting cotton apical buds in complex cotton field environments. Based on the YOLOv8 framework, the algorithm introduces Spatial and Channel Reconstruction Convolutions (SCConv) into the C2f module of the backbone network to reduce feature redundancy; embeds an Efficient Multi-scale Attention (EMA) module in the neck network to enhance feature extraction capabilities; and replaces the bounding box loss function with a dynamic non-monotonic focusing mechanism, WIoU, to accelerate model convergence. Experimental results on cotton apical bud data collected in complex field environments show that, compared to the original YOLOv8n algorithm, the YOLO-SEW algorithm reduces parameter count by 40.63%, computational load by 25%, and model size by 33.87%, while improving precision, recall, and mean average precision (mAP) by 1.2%, 2.5%, and 1.4%, respectively. Deployed on a Jetson Orin NX edge computing device and accelerated with TensorRT, the algorithm achieves a detection speed of 48 frames per second, effectively supporting real-time recognition of cotton apical buds and mechanized topping operations. Full article

Multimodal object detection is currently a research hotspot in computer vision. However, the fusion of visible and infrared modalities inevitably increases computational complexity, making most high-performance detection models difficult to deploy on resource-constrained UAV edge devices. Although pruning and knowledge distillation are widely used for model compression, applying them independently often leads to an unstable accuracy–efficiency trade-off. Therefore, this paper proposes a hybrid lightweight algorithm named SAMKD, which combines selective activation pruning with masked knowledge distillation in a staged manner to improve efficiency while maintaining detection performance. Specifically, the selective activation network pruning model (SAPM) first reduces redundant computation by dynamically adjusting network weights and the activation state of input data to generate a lightweight student network. Then, the mask binary classification knowledge distillation (MBKD) strategy is introduced to compensate for this degradation by guiding the student network to recover missing representation patterns under masked feature learning. Moreover, MBKD reformulates classification logits into multiple foreground–background binary mappings, effectively alleviating the severe foreground–background imbalance commonly observed in UAV aerial imagery. This paper constructs a multimodal UAV aerial imagery object detection dataset, M2UD-18K, which includes 9 types of targets and over 18,000 pairs. Extensive experiments show that SAMKD performs well on the self-constructed M2UD-18K dataset, as well as the public DroneVehicle dataset, achieving a favorable trade-off between detection accuracy and detection speed. Full article