Search Results (167)

To address the challenges of small target detection in UAV aerial images—such as difficulty in feature extraction, complex background interference, high miss rates, and stringent real-time requirements—this paper proposes an innovative model series named SRTSOD-YOLO, based on YOLO11. The backbone network incorporates a Multi-scale Feature Complementary Aggregation Module (MFCAM), designed to mitigate the loss of small target information as network depth increases. By integrating channel and spatial attention mechanisms with multi-scale convolutional feature extraction, MFCAM effectively locates small objects in the image. Furthermore, we introduce a novel neck architecture termed Gated Activation Convolutional Fusion Pyramid Network (GAC-FPN). This module enhances multi-scale feature fusion by emphasizing salient features while suppressing irrelevant background information. GAC-FPN employs three key strategies: adding a detection head with a small receptive field while removing the original largest one, leveraging large-scale features more effectively, and incorporating gated activation convolutional modules. To tackle the issue of positive-negative sample imbalance, we replace the conventional binary cross-entropy loss with an adaptive threshold focal loss in the detection head, accelerating network convergence. Additionally, to accommodate diverse application scenarios, we develop multiple versions of SRTSOD-YOLO by adjusting the width and depth of the network modules: a nano version (SRTSOD-YOLO-n), small (SRTSOD-YOLO-s), medium (SRTSOD-YOLO-m), and large (SRTSOD-YOLO-l). Experimental results on the VisDrone2019 and UAVDT datasets demonstrate that SRTSOD-YOLO-n improves the mAP@0.5 by 3.1% and 1.2% compared to YOLO11n, while SRTSOD-YOLO-l achieves gains of 7.9% and 3.3% over YOLO11l, respectively. Compared to other state-of-the-art methods, SRTSOD-YOLO-l attains the highest detection accuracy while maintaining real-time performance, underscoring the superiority of the proposed approach. Full article

The surface temperature of grain piles is sensitive to environmental fluctuations and exhibits nonlinear, multi-scale temporal patterns, making accurate prediction crucial for grain storage risk early warning. This paper proposes a decomposition–reconstruction prediction method integrating Sample Entropy (SampEn), variational mode decomposition (VMD), and a variant Long Short-Term Memory network (vLSTM). SampEn determines the optimal decomposition parameters, VMD extracts intrinsic mode functions (IMFs), and vLSTM, with peephole connections and coupled gates, conducts synchronous multi-IMF prediction. To explicitly account for environmental influences, a support vector regression (SVR) model driven by dew point temperature and vapor pressure deficit is employed to estimate the surface temperature variation $\mathrm{\Delta }T$. This component enhances the adaptability of the framework to dynamic storage conditions. The environment-derived $\mathrm{\Delta }T$ is then integrated with the VMD-SampEn-vLSTM output to obtain the final forecast. Experiments on real-granary data from Liaoning, China demonstrate that the proposed method reduces mean absolute error (MAE) and root mean square error (RMSE) by 25% and 14%, respectively, compared with baseline models, thus achieving a significant improvement in prediction performance. This integration of data-driven prediction with environmental adjustment significantly improves forecasting accuracy and robustness. Full article

Lithium-ion batteries (LIBs) are critical components in safety-critical systems such as electric vehicles, aerospace, and grid-scale energy storage. Their degradation over time can lead to catastrophic failures, including thermal runaway and uncontrolled combustion, posing severe threats to human safety and infrastructure. Developing a robust AI framework for degradation prognostics in safety-critical systems is essential to mitigate these risks and ensure operational safety. However, sensor noise, dynamic operating conditions, and the multi-scale nature of degradation processes complicate this task. Traditional denoising and modeling approaches often fail to preserve informative temporal features or capture both abrupt fluctuations and long-term trends simultaneously. To address these limitations, this paper proposes a hybrid data-driven framework that combines Sample Entropy-guided Variational Mode Decomposition (SE-VMD) with K-means clustering for adaptive signal preprocessing. The SE-VMD algorithm automatically determines the optimal number of decomposition modes, while K-means separates high- and low-frequency components, enabling robust feature extraction. A dual-branch architecture is designed, where Gated Recurrent Units (GRUs) extract short-term dynamics from high-frequency signals, and Transformers model long-term trends from low-frequency signals. This dual-branch approach ensures comprehensive multi-scale degradation feature learning. Additionally, experiments with varying sliding window sizes are conducted to optimize temporal modeling and enhance the framework’s robustness and generalization. Benchmark dataset evaluations demonstrate that the proposed method outperforms traditional approaches in prediction accuracy and stability under diverse conditions. The framework directly contributes to Artificial Intelligence for Security by providing a reliable solution for battery health monitoring in safety-critical applications, enabling early risk mitigation and ensuring operational safety in real-world scenarios. Full article

Grassland classification via remote sensing is essential for ecosystem monitoring and precision management, yet conventional satellite-based approaches are fundamentally constrained by coarse spatial resolution. To overcome this limitation, we harness high-resolution UAV multi-sensor data, integrating multi-scale image fusion with deep learning to achieve fine-scale grassland classification that satellites cannot provide. First, four categories of UAV data, including RGB, multispectral, thermal infrared, and LiDAR point cloud, were collected, and a fused image tensor consisting of 10 channels (NDVI, VCI, CHM, etc.) was constructed through orthorectification and resampling. For feature-level fusion, four deep fusion networks were designed. Among them, the MultiScale Pyramid Fusion Network, utilizing a pyramid pooling module, effectively integrated spectral and structural features, achieving optimal performance in all six image fusion evaluation metrics, including information entropy (6.84), spatial frequency (15.56), and mean gradient (12.54). Subsequently, training and validation datasets were constructed by integrating visual interpretation samples. Four backbone networks, including UNet++, DeepLabV3+, PSPNet, and FPN, were employed, and attention modules (SE, ECA, and CBAM) were introduced separately to form 12 model combinations. Results indicated that the UNet++ network combined with the SE attention module achieved the best segmentation performance on the validation set, with a mean Intersection over Union (mIoU) of 77.68%, overall accuracy (OA) of 86.98%, F1-score of 81.48%, and Kappa coefficient of 0.82. In the categories of Leymus chinensis and Puccinellia distans, producer’s accuracy (PA)/user’s accuracy (UA) reached 86.46%/82.30% and 82.40%/77.68%, respectively. Whole-image prediction validated the model’s coherent identification capability for patch boundaries. In conclusion, this study provides a systematic approach for integrating multi-source UAV remote sensing data and intelligent grassland interpretation, offering technical support for grassland ecological monitoring and resource assessment. Full article

With the rise in physiological data sampled from wearable devices, efficient methods must be developed to encode temporal information for the comparison of time series arising from uncontrolled monitoring. We present a fast, nonparametric method called Multiscale Average Absolute Difference (MSAAD) to extract multiscale temporal features from wearable device data for purposes ranging from statistical analysis to machine learning inference. MSAAD outperforms comparable algorithms like multiscale sample entropy (MSSE) and multiscale Katz Fractal Dimension (MS-KFD) in terms of calculation stability on short realizations and faster runtime. MSAAD outperforms MSSE and MS-KFD by being able to separate diabetic and non-diabetic cohorts with moderate and large effect sizes in both sexes. Furthermore, it is capable of capturing “critical slowing down” in the temperature dynamics of aging populations, a phenomenon that has been previously observed in controlled settings. We propose that MSAAD is a scalable, interpretable time series feature that is capable of identifying meaningful differences in physiological time series data without making assumptions regarding underlying process models. MSAAD could improve the ability to derive insight from time series data mining for health applications. Full article

Anomaly detection plays a vital role in ensuring product quality and operational safety across various industrial applications, from manufacturing to infrastructure monitoring. However, current methods often struggle with challenges such as limited generalization to complex multimodal anomalies, poor adaptation to domain-specific patterns, and reduced feature discriminability due to domain gaps between pre-trained models and industrial data. To address these issues, we propose AnomNet, a novel deep anomaly detection framework that integrates a lightweight feature adapter module to bridge domain discrepancies and enhance multi-scale feature discriminability from pre-trained backbones. AnomNet is trained using a dual-stage centroid learning strategy: the first stage employs separation and entropy regularization losses to stabilize and optimize the centroid representation of normal samples; the second stage introduces a centroid-based contrastive learning mechanism to refine decision boundaries by adaptively managing inter- and intra-class feature relationships. The experimental results on the MVTec AD dataset demonstrate the superior performance of AnomNet, achieving a 99.5% image-level AUROC and 98.3% pixel-level AUROC, underscoring its effectiveness and robustness for anomaly detection and localization in industrial environments. Full article

Electrical networks face operational challenges from power quality-affecting disturbances. Since disturbance signatures directly affect classifier performance, optimized feature selection becomes critical for accurate power quality assessment. The pursuit of robust feature extraction inevitably constrains the dimensionality of the discriminative feature set, but the complexity of the recognition model will be increased and the recognition speed will be reduced if the feature vector dimension is too high. Building upon the aforementioned requirements, in this paper, we propose a feature extraction framework that combines improved symplectic geometric mode decomposition, refined generalized multiscale quantum entropy, and refined generalized multiscale reverse dispersion entropy. Firstly, based on the intrinsic properties of power quality disturbance (PQD) signals, the embedding dimension of symplectic geometric mode decomposition and the adaptive mode component screening method are improved, and the PQD signal undergoes tri-band decomposition via improved symplectic geometric mode decomposition (ISGMD), yielding distinct high-frequency, medium-frequency, and low-frequency components. Secondly, utilizing the enhanced symplectic geometric mode decomposition as a foundation, the perturbation features are extracted by the combination of refined generalized multiscale quantum entropy and refined generalized multiscale reverse dispersion entropy to construct high-precision and low-dimensional feature vectors. Finally, a double-layer composite power quality disturbance model is constructed by a deep extreme learning machine algorithm to identify power quality disturbance signals. After analysis and comparison, the proposed method is found to be effective even in a strong noise environment with a single interference, and the average recognition accuracy across different noise environments is 97.3%. Under the complex conditions involving multiple types of mixed perturbations, the average recognition accuracy is maintained above 96%. Compared with the existing CNN + LSTM method, the recognition accuracy of the proposed method is improved by 3.7%. In addition, its recognition accuracy in scenarios with small data samples is significantly better than that of traditional methods, such as single CNN models and LSTM models. The experimental results show that the proposed strategy can accurately classify and identify various power quality interferences and that it is better than traditional methods in terms of classification accuracy and robustness. The experimental results of the simulation and measured data show that the combined feature extraction methodology reliably extracts discriminative feature vectors from PQD. The double-layer combined classification model can further enhance the model’s recognition capabilities. This method has high accuracy and certain noise resistance. In the 30 dB white noise environment, the average classification accuracy of the model is 99.10% for the simulation database containing 63 PQD types. Meanwhile, for the test data based on a hardware platform, the average accuracy is 99.03%, and the approach’s dependability is further evidenced by rigorous validation experiments. Full article

The hydropower unit is the core of the hydropower station, and maintaining the safety and stability of the hydropower unit is the first essential priority of the operation of the hydropower station. However, the complex environment increases the probability of the failure of hydropower units. Therefore, aiming at the complex diversity of hydropower unit faults and the imbalance of fault data, this paper proposes a fault identification method based on modified fractional-order hierarchical fluctuation dispersion entropy (MFHFDE) and AdaBoost-stochastic configuration networks (AdaBoost-SCN). First, the modified hierarchical entropy and fractional-order theory are incorporated into the multiscale fluctuation dispersion entropy (MFDE) to enhance the responsiveness of MFDE to various fault signals and address its limitation of overlooking the high-frequency components of signals. Subsequently, the Euclidean distance is used to select the fractional order. Then, a novel method for evaluating the complexity of time-series signals, called MFHFDE, is presented. In addition, the AdaBoost algorithm is used to integrate stochastic configuration networks (SCN) to establish the AdaBoost-SCN strong classifier, which overcomes the problem of the weak generalization ability of SCN under the condition of an unbalanced number of signal samples. Finally, the features extracted via MFHFDE are fed into the classifier to accomplish pattern recognition. The results show that this method is more robust and effective compared with other methods in the anti-noise experiment and the feature extraction experiment. In the six kinds of imbalanced experimental data, the recognition rate reaches more than 98%. Full article

Deep learning models have obvious advantages in detecting oil spills, but the training of deep learning models heavily depends on a large number of samples of high quality. However, due to the accidental nature, unpredictability, and urgency of oil spill incidents, it is difficult to obtain a large number of labeled samples in real oil spill monitoring scenarios. Surprisingly, few-shot learning can achieve excellent classification performance with only a small number of labeled samples. In this context, a new cross-domain few-shot SAR oil spill detection network is proposed in this paper. Significantly, the network is embedded with a hybrid attention feature extraction block, which consists of a coordinate attention module to perceive the channel information and spatial location information, as well as a global self-attention transformer module capturing the global dependencies and a multi-scale self-attention module depicting the local detailed features, thereby achieving deep mining and accurate characterization of image features. In addition, to address the problem that it is difficult to distinguish between the suspected oil film in seawater and real oil film using few-shot due to the small difference in features, this paper proposes a double loss function category determination block, which consists of two parts: a well-designed category-perception loss function and a traditional cross-entropy loss function. The category-perception loss function optimizes the spatial distribution of sample features by shortening the distance between similar samples while expanding the distance between different samples. By combining the category-perception loss function with the cross-entropy loss function, the network’s performance in discriminating between real and suspected oil films is thus maximized. The experimental results effectively demonstrate that this study provides an effective solution for high-precision oil spill detection under few-shot conditions, which is conducive to the rapid identification of oil spill accidents. Full article

Marine oil spills cause significant environmental damage worldwide. Marine radar imagery is used for oil spill detection. However, the rapid attenuation of backscatter intensity with increasing distance limits detectable coverage. A multi-stage image enhancement framework integrating background clutter fitting subtraction, Multi-Scale Retinex, and Gamma correction is proposed. Experimental results using marine radar images sampled in the oil spill incident in Dalian 2010 are used to demonstrate the significant improvements. Compared to Contrast-Limited Adaptive Histogram Equalization and Partially Overlapped Sub-block Histogram Equalization, the proposed method enhances image contrast by 24.01% and improves the measurement of enhancement by entropy by 17.11%. Quantitative analysis demonstrates 95% oil spill detection accuracy through visual interpretation, while significantly expanding detectable coverage for oil extraction. Full article

Coal mine disaster safety monitoring often employs microseismic technology for its high sensitivity and real-time capability. However, nonlinear, non-stationary, and multi-scale signals limit traditional time series models (e.g., ARMA, ARIMA). This paper proposes a hybrid deep learning model—VMD-SE-CEEMDAN-BO-CNNGRU—integrating variational mode decomposition, sample entropy, CEEMDAN, Bayesian optimization, and a CNN-GRU architecture. Microseismic data from the 08 working face in D mine (Weibei mining area) were used to predict daily maximum energy, average energy, and frequency. The model achieved high predictive performance with R² values of 0.93, 0.89, and 0.88, significantly outperforming baseline models lacking modal decomposition. Comparative experiments verified the superiority of the VMD-first, SE-reconstruction, and CEEMDAN-second decomposition strategy, yielding up to 13% greater accuracy than reverse-order schemes. The model maintained R² above 0.80 on another dataset from the 03 working face in W mine (Binchang mining area), demonstrating robust generalization. Although performance declined during fault disturbances, accuracy for average energy and frequency rebounded post-disturbance, indicating strong adaptability. Overall, the VSCB-CNNGRU model enhances both accuracy and stability in microseismic prediction, supporting dynamic risk assessment and early warning in coal mining. Full article

Dissolved Gas Analysis (DGA) of transformer oil is a critical technique for transformer fault diagnosis that involves analyzing the concentration of gases to detect potential transformer faults in a timely manner. Given the issues of large model parameters and high computational resource demands in transformer DGA diagnostics, this study proposes a lightweight convolutional neural network (CNN) model for improving gas ratio methods, combining Knowledge Distillation (KD) and recursive plots. The approach begins by extracting features from DGA data using the ratio method and Multiscale sample entropy (MSE), then reconstructs the state space of the feature data using recursive plots to generate interpretable two-dimensional image features. A deep feature extraction process is performed using the ResNet50 model, integrated with the Convolutional Block Attention Module (CBAM). Subsequently, the Sparrow Optimization Algorithm (SSA) is applied to optimize the hyperparameters of the ResNet50 model, which is trained on DGA data as the teacher model. Finally, a dual-path distillation mechanism is introduced to transfer the efficient features and knowledge from the teacher model to the student model, MobileNetV3-Large. The experimental results show that the distilled model reduces memory usage by 83.5% and computation time by 73.2%, significantly lowering computational complexity while achieving favorable performance across various evaluation metrics. This provides a novel technical solution for the improvement of gas ratio methods. Full article

Surface cracks serve as early warning signals for potential geological hazards, and their precise segmentation is crucial for disaster risk assessment. Due to differences in acquisition conditions and the diversity of crack morphology, scale, and surface texture, there is a significant domain shift between different crack datasets, necessitating transfer training. However, in real work areas, the sparse distribution of cracks results in a limited number of samples, and the difficulty of crack annotation makes it highly inefficient to use a high proportion of annotated samples for transfer training to predict the remaining samples. Domain adaptation methods can achieve transfer training without relying on manual annotation, but traditional domain adaptation methods struggle to effectively address the characteristics of cracks. To address this issue, we propose an unsupervised domain adaptation method for crack segmentation. By employing a hierarchical adversarial mechanism and a prediction entropy minimization constraint, we extract domain-invariant features in a multi-scale feature space and sharpen decision boundaries. Additionally, by integrating a Mix-Transformer encoder, a multi-scale dilated attention module, and a mixed convolutional attention decoder, we effectively solve the challenges of cross-domain data distribution differences and complex scene crack segmentation. Experimental results show that UCrack-DA achieves superior performance compared to existing methods on both the Roboflow-Crack and UAV-Crack datasets, with significant improvements in metrics such as mIoU, mPA, and Accuracy. In UAV images captured in field scenarios, the model demonstrates excellent segmentation Accuracy for multi-scale and multi-morphology cracks, validating its practical application value in geological hazard monitoring. Full article

In recent years, the continuous development of deep learning has significantly advanced its application in the field of remote sensing. However, the semantic segmentation of high-resolution remote sensing images remains challenging due to the presence of multi-scale objects and intricate spatial details, often leading to the loss of critical information during segmentation. To address this issue and enable fast and accurate segmentation of remote sensing images, we made improvements based on SegNet and named the enhanced model CSNet. CSNet is built upon the SegNet architecture and incorporates a coordinate attention (CA) mechanism, which enables the network to focus on salient features and capture global spatial information, thereby improving segmentation accuracy and facilitating the recovery of spatial structures. Furthermore, skip connections are introduced between the encoder and decoder to directly transfer low-level features to the decoder. This promotes the fusion of semantic information at different levels, enhances the recovery of fine-grained details, and optimizes the gradient flow during training, effectively mitigating the vanishing gradient problem and improving training efficiency. Additionally, a hybrid loss function combining weighted cross-entropy and Dice loss is employed. To address the issue of class imbalance, several categories within the dataset are merged, and samples with an excessively high proportion of background pixels are removed. These strategies significantly enhance the segmentation performance, particularly for small-sample classes. Experimental results from the Five-Billion-Pixels dataset demonstrate that, while introducing only a modest increase in parameters compared to SegNet, CSNet achieves superior segmentation performance in terms of overall classification accuracy, boundary delineation, and detail preservation, outperforming established methods such as U-Net, FCN, DeepLabv3+, SegNet, ViT, HRNe and BiFormert. Full article

Monitoring the blast furnace shaft static pressure is crucial for maintaining a stable ironmaking process. Traditional rule-based methods and manual inspections suffer from high labor costs and inconsistent standards. This article proposes a new unsupervised anomaly detection framework that combines adversarial autoencoder with variational mode decomposition (VMD). Firstly, using VMD combined with sample entropy calculation and clustering algorithm, the trend, period, and other components of multidimensional signals are extracted, and then these components are integrated into an improved adversarial training autoencoder to detect global and local anomalies. The proposed method has an accuracy of 0.95, a recall rate of 0.91, and an F1 score of 0.93. Which demonstrates the method effectively captures multi-scale anomalies including value bias, morphological changes, and sudden fluctuations, while providing analysts with interpretable anomaly detail diagnosis. Full article