Search Results (9,943)

The open-vocabulary understanding of UAV aerial images plays a crucial role in enhancing the intelligence level of remote sensing applications, such as disaster assessment, precision agriculture, and urban planning. In this paper, we propose an innovative open-vocabulary model for UAV images, which combines vision–language methods to achieve efficient recognition and segmentation of unseen categories by generating multi-view image descriptions and feature extraction. To enhance the generalization ability and robustness of the model, we adopted Mixup technology to blend multiple UAV images, generating more diverse and representative training data. To address the limitations of existing open-vocabulary models in UAV image analysis, we leverage the GPT model to generate accurate and professional text descriptions of aerial images, ensuring contextual relevance and precision. The image encoder utilizes a U-Net with Mamba architecture to extract key point information through edge detection and partition pooling, further improving the effectiveness of feature representation. The text encoder employs a fine-tuned BERT model to convert text descriptions of UAV images into feature vectors. Three key loss functions were designed: Generalization Loss to balance old and new category scores, semantic segmentation loss to evaluate model performance on UAV image segmentation tasks, and Triplet Loss to enhance the model’s ability to distinguish features. The Comprehensive Loss Function integrates these terms to ensure robust performance in complex UAV segmentation tasks. Experimental results demonstrate that the proposed method has significant advantages in handling unseen categories and achieving high accuracy in UAV image segmentation tasks, showcasing its potential for practical applications in diverse aerial imagery scenarios. Full article

This study explores a method of detecting smoke plumes effectively as the early sign of a forest fire. Convolutional neural networks (CNNs) have been widely used for forest fire detection; however, they have not been customized or optimized for smoke characteristics. This paper proposes a CNN-based forest smoke detection model featuring novel backbone architecture that can increase detection accuracy and reduce computational load. Since the proposed backbone detects the plume of smoke through different views using kernels of varying sizes, it can better detect smoke plumes of different sizes. By decomposing the traditional square kernel convolution into a depth-wise convolution of the coordinate kernel, it can not only better extract the features of the smoke plume spreading along the vertical dimension but also reduce the computational load. An attention mechanism was applied to allow the model to focus on important information while suppressing less relevant information. The experimental results show that our model outperforms other popular ones by achieving detection accuracy of up to 52.9 average precision (AP) and significantly reduces the number of parameters and giga floating-point operations (GFLOPs) compared to the popular models. Full article

Magnetotelluric (MT) forward modeling is a key technique in magnetotelluric sounding, and deep learning has been widely applied to MT forward modeling. In three-dimensional (3-D) problems, although existing methods can predict forward modeling results with high accuracy, they often use multiple networks to simulate multiple forward modeling parameters, resulting in low efficiency. We apply multi-task learning (MTL) to 3-D MT forward modeling to achieve simultaneous inference of apparent resistivity and impedance phase, effectively improving overall efficiency. Furthermore, through comparative analysis of feature map differences in various decoder layers of the network, we identify the optimal branching point for multi-task learning decoders. This enhances the feature extraction capabilities of the network and improves the prediction accuracy of forward modeling parameters. Additionally, we introduce an uncertainty-based loss function to dynamically balance the learning weights between tasks, addressing the shortcomings of traditional loss functions. Experiments demonstrate that compared with single-task networks and existing multi-task networks, the proposed network (MT-FeatureNet) achieves the best results in terms of Structural Similarity Index Measure (SSIM), Mean Relative Error (MRE), and Mean Absolute Error (MAE). The proposed multi-task learning model not only improves the efficiency and accuracy of 3-D MT forward modeling but also provides a novel approach to the design of multi-task learning network structures. Full article

An on-site survey is the primary task of working live in distribution networks. However, the traditional manual on-site survey method is not only not very intuitive but also inefficient. The application of 3D point cloud technology has opened up new avenues for on-site surveys in life working in distribution networks. This paper focused on the application of the Point Transformer V2(PTV2) model in the segmentation of distribution network point clouds. Given its deficiencies in boundary discrimination ability and limited feature extraction ability when processing the point clouds of distribution networks, an improved Non-local Focal Loss-Point Transformer V2 (NF-PTV2) model was proposed. With PTV2 as its core, this model incorporated the Non-Local attention to capturing long-distance feature dependencies, thereby compensating for the PTV2 model’s shortcomings in extracting features of large-scale objects with complex features. Simultaneously, the Focal Loss function was introduced to address the issue of class imbalance and enhance the model’s learning ability for small complex samples. The experimental results demonstrated that the overall accuracy (OA) of this model on the distribution network dataset reached 93.28%, the mean intersection over union (mIoU) reached 81.58%, and the mean accuracy (mAcc) reached 87.21%. In summary, the NF-PTV2 model proposed in this article demonstrated good performance in the point cloud segmentation task of the distribution network and can accurately identify various objects, which, to some extent, overcomes the limitations of the PTV2 model. Full article

Wave attenuation by natural coastal features is recognised as a soft engineering approach to shoreline protection from storm surges and destructive waves. The effectiveness of wave energy dissipation is determined, in part, by vegetation structure, extent, and distribution. Mangroves line ca. 15% of the world’s coastlines, primarily in tropical and subtropical regions but also extending into temperate climates, where mangroves are shorter and multi-stemmed. Using wave loggers deployed across mangrove and non-mangrove shorelines, we studied the wave attenuating capacity and the drag coefficient (C_D) of temperate Avicennia marina mangrove forests of varying structure in Western Port, Australia. The structure of the vegetation obstructing the flow path was represented along each transect in a three-dimensional point cloud derived from overlapping uncrewed aerial vehicle (UAV) images and structure-from-motion (SfM) algorithms. The wave attenuation coefficient (b) calculated from a fitted exponential decay model at the vegetated sites was on average 0.011 m⁻¹ relative to only 0.009 m⁻¹ at the unvegetated site. We calculated a C_D for this forest type that ranged between 2.7 and 4.9, which is within the range of other pencil-rooted species such as Sonneratia sp. but significantly lower than prop-rooted species such as Rhizophora spp. Wave attenuation efficiency significantly decreased with increasing water depth, highlighting the dominance of near-bed friction on attenuation in this forest type. The UAV-derived point cloud did not describe the vegetation (especially near-bed) in sufficient detail to accurately depict the obstacles. We found that a temperate mangrove greenbelt of just 100 m can decrease incoming wave heights by close to 70%, indicating that, similarly to tropical and subtropical forests, temperate mangroves significantly attenuate incoming wave energy under normal sea conditions. Full article

Electricity theft, emerging as one of the severe cyberattacks in smart grids, causes significant economic losses. Due to the powerful expressive ability of deep neural networks (DNN), supervised and unsupervised DNN-based electricity theft detection (ETD) schemes have experienced widespread deployment. However, existing works have the following weak points: Supervised DNN-based schemes require abundant labeled anomalous samples for training, and even worse, cannot detect unseen theft patterns. To avoid the extensively labor-consuming activity of labeling anomalous samples, unsupervised DNNs-based schemes aim to learn the normality of time-series and infer an anomaly score for each data instance, but they fail to capture periodic features effectively. To address these challenges, this paper proposes a novel periodicity-enhanced deep one-class classification framework (PE-DOCC) based on a periodicity-enhanced transformer encoder, named Periodicformer encoder. Specifically, within the encoder, a novel criss-cross periodic attention is proposed to capture both horizontal and vertical periodic features. The Periodicformer encoder is pre-trained by reconstructing partially masked input sequences, and the learned latent representations are then fed into a one-class classification for anomaly detection. Extensive experiments on real-world datasets demonstrate that our proposed PE-DOCC framework outperforms state-of-the-art unsupervised ETD methods. Full article

With ever-increasing incidence and high metastatic potential, cutaneous melanoma is the deadliest skin cancer. Risk prediction based on the Tumor-Node-Metastasis (TNM) staging system has medium accuracy with intermediate IIB-IIIB stages, as roughly 25% of patients with low-medium-grade TNM, and hence a favorable prognostic, undergo an aggressive disease with short survival and around 15% of deaths arise from metastases of thin, low-risk lesions. Therefore, reliable prognostic biomarkers are required. We used genomic and clinical information of melanoma patients from the TCGA-SKCM cohort and two GEO studies for discovery and validation of potential biomarkers, respectively. Neither mutation nor overexpression of major melanoma driver genes provided significant prognostic information. Conversely, expression of MGRN1 and the melanocyte-specific genes MLANA, PMEL, and TYRP1 provided a simple 4-gene signature identifying with high-sensitivity (>80%), low-medium TNM patients with adverse outcomes. Transcriptomic analysis of tumors with this signature, or from low-medium-grade TNM patients with poor outcomes, revealed comparable dysregulation of an inflammatory response, cell cycle progression, and DNA damage/repair programs. A functional analysis of MGRN1-knockout cells confirmed these molecular features. Therefore, the simple MGRN1-MLANA-PMEL-TYRP1 combination of biomarkers complemented TNM staging prognostic accuracy and pointed to the dysregulation of immunological responses and genomic stability as determinants of a melanoma outcome. Full article

Landscape visual evaluation is a key method for assessing the value of visual landscape resources. This study aims to enhance the visual environment and sensory quality of urban landscapes by establishing standards for the visual comfort of urban natural landscapes. Using line-of-sight and multi-factor analysis algorithms, the method assesses spatial visibility and visual exposure of building clusters in the core urban areas of Harbin, identifying areas and viewpoints with high visual potential. Focusing on the viewpoints of landmark 3D models and the surrounding landscape’s visual environment, the study uses the city’s sky, greenery, and water features as key visual elements for evaluating the comfort of urban natural landscapes. By integrating GIS data, big data street-view photos, and image semantic recognition, spatial analysis algorithms extract both objective and subjective visual values at observation points, followed by mathematical modeling and quantitative analysis. The study explores the coupling relationship between objective physical visual values and subjective perceived visibility. The results show that 3D visual analysis effectively reveals the relationship between landmark buildings and surrounding landscapes, providing scientific support for urban planning and contributing to the development of a more distinctive and attractive urban space. Full article

The presence of non-tobacco-related materials can significantly compromise the quality of tobacco. To accurately detect non-tobacco-related materials, this study introduces a lightweight and real-time detection model derived from the YOLOv11 framework, named LRNTRM-YOLO. Initially, due to the sub-optimal accuracy in detecting diminutive non-tobacco-related materials, the model was augmented by incorporating an additional layer dedicated to enhancing the detection of small targets, thereby improving the overall accuracy. Furthermore, an attention mechanism was incorporated into the backbone network to focus on the features of the detection targets, thereby improving the detection efficacy of the model. Simultaneously, for the introduction of the SIoU loss function, the angular vector between the bounding box regressions was utilized to define the loss function, thus improving the training efficiency of the model. Following these enhancements, a channel pruning technique was employed to streamline the network, which not only reduced the parameter count but also expedited the inference process, yielding a more compact model for non-tobacco-related material detection. The experimental results on the NTRM dataset indicate that the LRNTRM-YOLO model achieved a mean average precision (mAP) of 92.9%, surpassing the baseline model by a margin of 4.8%. Additionally, there was a 68.3% reduction in the parameters and a 15.9% decrease in floating-point operations compared to the baseline model. Comparative analysis with prominent models confirmed the superiority of the proposed model in terms of its lightweight architecture, high accuracy, and real-time capabilities, thereby offering an innovative and practical solution for detecting non-tobacco-related materials in the future. Full article

Background/Objectives: Pleural effusion represents an accumulation of fluid in the pleural cavity, frequently associated with pneumonia. There has been a gradual increase in cases among children in recent years, with a notable rise during the post-pandemic period, potentially due to immune debt, decreased vaccination coverage, and changes in pathogen dynamics. Methods: We enrolled 66 children with pleural effusion treated at the Children’s Emergency Clinical Hospital, Brasov, between January 2019 and September 2024. We analyzed the data on demographics, symptoms, vaccination status, hospitalization, and treatments to assess the trends in the incidence and clinical features. Results: The median age was 5 years (ranging from 3 months to 17 years). Most patients were male (57.5%) from rural areas (34.8%). Only 40.9% fulfilled the vaccination schedule of Romania. We observed a rise in hospitalizations in the last two years, with 16 cases in 2023 and 15 in 2024, and most were being admitted in April (15.5%). Patients mainly had severe (36%) and medium (26%) acute respiratory failure. S. pneumoniae was the most common isolate with two cases each of serotype 1, 14, and 23A, and one case each of serotype 3, 31, and 34, followed by H. influenzae and P. aeruginosa. Treatment was mostly with ceftriaxone (69.6%), Vancomycin (63.6%), Meropenem (53.0%), and Teicoplanin (25.7%). Some children required thoracic drainage (34.8%). Complications like pneumothorax (16.6%), polyserositis (4.5%), and pneumomediastinum (3.0%) were found. Conclusions: The rise in pleural effusion cases may be influenced by various factors, such as changes in pathogen behavior or host immune responses following the pandemic. Further research is needed to understand these potential mechanisms. The emergence of non-PCV20 strains and the common occurrence of serotype 3 infections point out the need to study serotype trends and evaluate whether expanding vaccine programs could be beneficial. Full article

Continuous spalling exposure can weaken the performance of structures. Therefore, the development of methods for detecting wall spall damage remains essential in the field of Structural Health Monitoring. Currently, researchers mainly rely on 2D information for spall detection and predominantly use manual data collection methods in the complex environment of residential buildings, which are usually inefficient. To address this challenge, an automated 3D image processing system for wall spalls is proposed in this study. First, UGV path planning was performed in order to collect information about the surrounding environmental defects. Second, to address the shortcomings of RandLA-Net, a dynamic enhanced dual-branch structure is established based on which consistency constraints are introduced, a lightweight attention module is added, and the loss function is optimized in order to enhance the ability of the model in extracting feature information of the point cloud. Finally, spalls are quantitatively evaluated to determine the damage to buildings. The results show that the Randla-Spall achieves 94.71% Recall and 84.20% mIoU on the test set, improved by 4.25% and 5.37%. An integrated process using a lightweight device is achieved in this study, which is capable of efficiently extracting and quantifying spalling defects and provides valuable references for SHM. Full article

South Africa is a leader in the scholarship on green urbanism in the Global South, but academic progress has not translated to broad implementation. Notably, government-subsidized housing projects have produced peripheral developments featuring low build quality, conventional gray infrastructure, and deficient socio-economic and environmental amenities. Declining delivery and increasing informal settlement spawned a 2014 shift to housing megaprojects to increase output and improve living conditions, socio-economic integration, and sustainability. The shift offered opportunities for a normative focus on greener development mirrored in the discourse surrounding project descriptions. Yet, the level of enactment has remained unclear. In reflecting on these points, this paper employs environmental justice as a theoretical framework and completes a comprehensive review of the academic literature on housing megaprojects and the depth of their greener development commitments. A three-phase, seven-stage review protocol retrieves the relevant literature, and bibliometric and qualitative content analyses identify publication trends and themes. Results indicate limited scholarship on new megaprojects with sporadic and superficial references to greener development, mostly reserved for higher-income segments and private developments. In response, this paper calls for more determined action to launch context-aware and just greener megaprojects and offers corresponding guidance for research and practice of value to South Africa and beyond. Full article

The present work proposes a new torque/speed equilibrium point monitoring technique for an aircraft Hybrid Electric Propulsion System (HEPS) through an accelerometric-signal-based approach. Sampled signals were processed using statistical indexes, filtering, and a feature reduction and selection algorithm to train a classification Feedforward Neural Network. A supervised Machine Learning model was developed to classify the HEPS operating modes characterized by an Internal Combustion Engine as a single propulsor or by combining the latter with an Electric Machine used as a motor or a generator. The abnormal changes in the torque/speed equilibrium point were detected by the monitoring index built by computing the Root Mean Square on the value identified by the classifier. The procedure was validated through experimental tests that demonstrated its validity. Full article

Point feature cartographic label placement is a key problem in the automatic configuration of map labeling. Prior research on it only addresses label conflict or overlap issues; it does not fully take into account and resolve both types of issues. In this study, we attempt to apply machine learning techniques to the automatic placement of point feature labels since label placement is a task that heavily relies on expert expertise, which is very congruent with neural networks’ ability to mimic the human brain’s thought process. We trained ResNet using large amounts of well-labeled picture data. The label’s proper location for a given unlabeled point feature was then predicted by the trained model. We assessed the outcomes both quantitatively and qualitatively, contrasting the ResNet model’s output with that of the expert manual placement approach and the conventional Maplex automatic placement method. According to the evaluation, the ResNet model’s test set accuracy was 97.08%, demonstrating its ability to locate the point feature label in the right place. This study offers a workable solution to the label overlap and conflict issue. Simultaneously, it has significantly enhanced the map’s esthetic appeal and the information’s clarity. Full article

Aerial optical cameras are the primary method for capturing high-resolution images to produce large-scale mapping products. To improve aerial photography efficiency, multiple cameras are often used in combination to generate large-format virtual central projection images. This paper presents a high-precision method for directly transforming raw images obtained from a dual-camera system mounted at an oblique angle into virtual central projection images, thereby enabling the construction of low-cost, large-format aerial camera systems. The method commences with an adaptive sub-block in the overlapping regions of the raw images to extract evenly distributed feature points, followed by iterative relative orientation to improve accuracy and reliability. A global projection transformation matrix is constructed, and the sigmoid function is employed as a weighted distance function for image stitching. The results demonstrate that the proposed method produces more evenly distributed feature points, higher relative orientation accuracy, and greater reliability. Simulation analysis of image overlap indicates that when the overlap exceeds 7%, stitching accuracy can be better than 1.25 μm. The aerial triangulation results demonstrate that the virtual central projection images satisfy the criteria for the production of 1:1000 scale mapping products. Full article