Accurate and reliable photovoltaic (PV) power forecasting is essential for the secure and economical operation of microgrids. Traditional point forecasting methods fail to represent predictive uncertainty, while interval forecasting is difficult to integrate into deterministic control and dispatch frameworks. To overcome these limitations, this paper proposes a Point–Interval Fusion Model (PIFM) that achieves joint optimization of accuracy and reliability within a unified framework. The proposed model employs a CNN-SE-BiLSTM network to extract spatiotemporal features and utilizes quantile regression to generate probabilistic prediction intervals. Moreover, an adaptive weighting mechanism is introduced to dynamically fuse the point forecast and the interval median according to interval width, maintaining an optimal trade-off between accuracy and robustness. Experiments conducted on a real-world PV plant dataset demonstrate that PIFM outperforms baseline models, reducing RMSE and MAE to 2.691 and 1.573, respectively, and achieving a PICP of 0.90 at the 90% confidence level. These results confirm that PIFM delivers both high-precision forecasts and reliable uncertainty quantification, providing enhanced decision support for microgrid control and scheduling.

Accurate detection of delamination in building facades is critical for prolonging service life and ensuring structural safety. Current inspection methodologies heavily rely on manual interpretation, lacking efficiency and intelligent robustness. While infrared thermography provides a non-destructive means for detecting subsurface delamination, its accuracy is often compromised by low thermal contrast under uncontrolled conditions and the absence of uncertainty quantification in deep learning models. To address these limitations, this paper proposes TIHSNet, a novel delamination detection framework based on semantic segmentation and uncertainty quantification. Specifically, a physics-informed thermal gradient attention module is introduced to emphasize thermodynamically meaningful gradients and enable accurate delamination boundary delineation. Subsequently, a dual output mechanism is proposed to simultaneously generate prediction and uncertainty maps, enabling quantitative assessment of predictive reliability and identification of regions requiring expert review. To further enhance spatial localization, visible light images are integrated to capture tile boundary information and support spatial classification of delamination. Experiments were conducted on a self constructed dataset comprising 2102 infrared thermography and visible light images collected from reinforced concrete and brick masonry walls. The results demonstrate that TIHSNet achieves a precision of 96.1%, surpassing traditional thresholding methods with a 27.9% gain, and further outperforming existing deep learning approaches by 10.5%. The uncertainty quantification results further validate the model’s robustness and its ability to support reliable decision making in real world inspection scenarios.

Extracting continuous road networks from highresolution remote sensing imagery remains challenging due to occlusions caused by tree canopies, shadows, and complex backgrounds. Although recent Transformer-based methods excel at modeling global dependencies, they often neglect the feature misalignment issues inherent in standard upsampling operators, leading to blurred boundaries and disconnected paths. To address these limitations, we propose PVTCARAFE, a novel road extraction network following a “global modeling-detail enhancement” philosophy. Specifically, the network utilizes a Pyramid Vision Transformer (PVT) backbone to capture robust semantic features. In the decoding stage, we introduce a Ridge module combined with Content-Aware ReAssembly of Features (CARAFE) to effectively align global semantics with local details, overcoming the detail loss common in traditional interpolation. Furthermore, a parameter-free SimAM attention mechanism is integrated to adaptively highlight narrow and weak road regions without significantly increasing model complexity. Extensive experiments on the Massachusetts and DeepGlobe datasets demonstrate that PVTCARAFE outperforms state-of-the-art methods in terms of accuracy and boundary connectivity, achieving a superior trade-off between performance and computational efficiency.

Recent advances in Mamba-based architectures have demonstrated promising potential for hyperspectral image classification (HSIC), offering linear-complexity long-range dependency modeling. However, two critical challenges persist in adapting this paradigm to hyperspectral image (HSI) analysis: the substantial requirement for annotated training samples and insufficient capacity to interpret the intricate spatial-spectral features inherent in HSI data, particularly under few-shot learning (FSL) scenarios. To address these limitations, we present SSupMamba, a novel self-supervised Mamba framework tailored for HSIC. First, we propose a composite scanning Mamba block (CSMB) that enables comprehensive global feature extraction through multidirectional selective scanning of HSI data cubes. Second, we develop a spatial-spectral masked Mamba (SAEM) framework that uses randomized masking and reconstruction tasks to enhance local representation learning. Third, we establish a unified self-supervised architecture incorporating contrastive learning (CL) to maximize mutual information between multiviews while preserving intrinsic spatial-spectral characteristics. Experimental results on four public datasets demonstrate that the proposed method exhibits excellent feature extraction capabilities under few-shot conditions and outperforms several state-of-the-art HSIC methods. The code is available at: https://github.com/Winkness/SSupMamba.

In the assembly of peg-hole composite components, traditional fixed-parameter admittance control struggles to adapt to varying contact conditions, leading to a high failure rate. To address this issue, this paper proposes a variable-parameter admittance control algorithm based on MADDPG with a segmented reward mechanism. This approach dynamically adjusts admittance parameters via reinforcement learning to enhance multi-robot collaborative assembly performance. Experiments were conducted on assemblies with three different tolerances (1mm, 2mm, and 3mm) and compared against fixed-parameter admittance control. Results demonstrate that the proposed method achieves rapid convergence across different tolerance levels and successfully completes the 2mm tolerance assembly within 100 steps, with a success rate exceeding 60%, significantly outperforming the fixed-parameter approach. The segmented reward mechanism effectively mitigates excessive contact forces and positional deviations, improving assembly stability. This research contributes to autonomous assembly in unstructured environments and has the potential to enhance robotic adaptability and reliability in industrial manufacturing.

Thermal comfort is a subjective state of consciousness in which a person expresses satisfaction with the thermal environment, determined by a combination of physiological and psychological factors. As a visual representation of psychological responses, body movements provide a new means for accurate assessment of thermal comfort. Aiming at the problem that the mapping relationship between body movements and thermal comfort is not yet clear, this paper carries out the human thermal comfort assessment experiment based on body movements, and proposes the universal quantification of thermal sensation and the personalized prediction model of thermal expression based on body movements. First, the categories of thermally uncomfortable body movements were clarified through questionnaire research, and the probabilistic statistical strategy was used to quantify the degree of representation of thermal sensation by these movements in a pervasive way, which was defined as the thermal discomfort behavior index (TDI); subsequently, experimental studies were conducted to explore the correlation between physiological characteristics and TDI, and to determine the key influencing factors of TDI; finally, the TDI index was combined with different combinations of physiological and environmental parameters, as well as the effects of multiple machine learning models in thermal comfort prediction, the optimal way of combining the TDI index with other parameters and machine learning models is determined through cross-validation methods, and a personalized thermal expression prediction model based on body movements is constructed. Through experimental validation, the model shows high accuracy and reliability, providing a new method for contactless thermal comfort environment control.

Aiming at the problem of low detection accuracy for driver fatigue in low-light environments, the CHDS-YOLO low-light fatigued driving detection algorithm based on the improved YOLOv9 is proposed. Firstly, lightweight improvements are made by replacing some conventional convolutions in the backbone network with Depthwise Separable Convolution (DSConv) to reduce the computational complexity of the model. In order to maintain performance during the lightweighting process, deformable convolutions are introduced to improve the RepNCSPELAN4 module, and the Reparameterized Deformable Convolutional Network (Rep_DCN) module is designed to enhance the feature extraction ability of the backbone network, so as to better handle detection targets with shape variations. Secondly, the Cross Hybrid Attention Fusion (CHAF) module is proposed. Through the cross structure, it can fuse the high-level and low-level features of images, highlight image details and alleviate the low-light problem. Finally, a visual channel attention mechanism is introduced into the detection neck. By combining global information with local information, it can capture the information relationships among different channels and selectively strengthen useful feature information. Experiments show that on the dataset of real scenes, the CHDS-YOLO algorithm proposed in this paper improves the mAP@0.5 by 1.6% and the mAP@0.5:0.95 by 1.5% compared with the original YOLOv9n algorithm. Meanwhile, it increases the FPS by 5.8% and reduces the number of model parameters by 13.3%. The accuracy of detecting driving fatigue in low-light environments and the lightweight design make it highly feasible in practical deployment.

For fine-grained recognition, capturing distinguishable features and effectively utilizing local information play a key role, since the objects of recognition exhibit subtle differences in different subcategories. Finding subtle differences between subclasses is not straightforward. To address this problem, we propose a weakly supervised fine-grained classification network model with Local Diversity Guidance (LDGNet). We designed a Multi-Attention Semantic Fusion Module (MASF) to build multi-layer attention maps and channel–spatial interaction, which can effectively enhance the semantic representation of the attention maps. We also introduce a random selection strategy (RSS) that forces the network to learn more comprehensive and detailed information and more local features from the attention map by designing three feature extraction operations. Finally, both the attention map obtained by RSS and the feature map are employed for prediction through a fully connected layer. At the same time, a dataset of ancient towers is established, and our method is applied to ancient building recognition for practical applications of fine-grained image classification tasks in natural scenes. Extensive experiments conducted on four fine-grained datasets and explainable visualization demonstrate that the LDGNet can effectively enhance discriminative region localization and detailed feature acquisition for fine-grained objects, achieving competitive performance over other state-of-the-art algorithms.

Accurate assessment of individual metabolic rate (MET) is essential for creating a comfortable indoor environment, computer vision offers a non-contact approach for predicting MET. Existing methods primarily identify activity types and assign MET based on established standards. However, these methods face limitations in predicting MET for activities beyond predefined categories and overlook individual differences, thus restricting prediction accuracy and applicability. To address these issues, this study proposes a visual prediction method for MET that integrates individual movement speed and physiological characteristics, allowing the prediction of MET for various activities performed by different individuals, and provide technical support for the personalized thermal comfort control. Specifically, visual technology is first applied to detect and track skeleton keypoints during an individual’s movement. These keypoints’ coordinates are used to estimate the individual’s height, followed by the calculation of body weight and surface area. Next, the dynamic changes in skeleton keypoints between frames are analyzed to calculate movement speed, and energy expenditure rate (EER) is estimated based on principles of kinetic energy for various movement states. Finally, the MET is calculated by dividing EER by body surface area. We predicted the MET of 20 subjects under seven different activities and recorded heart rate data throughout the experiment, and validated the MET calculated using the heart rates by comparing them with the predicted values. The average MAE for MET prediction was 0.35 met and the average RMSE was 0.43 met, indicating that the method can accurately predict the MET of different individuals when performing various activities.

Automatic road extraction from satellite images is a popular research hotspot topic in the field of remote sensing. However, the complexity of road features and structures hinders their precise extraction. Road features have different saliency due to the occlusion of other objects and diverse road types. Road structures are scattered across various locations and affected by geographical conditions, population distribution, and transportation demands. These issues render current methods based on convolutional neural networks prone to fragmented and incomplete road extraction results. To address these issues, a Road Extraction Network with Dual-View Information Perception Based on GCN (RDPGNet) is proposed, where the GCN-based dual-view perceptron (GDVP) uses the superior information interaction capability of the GCN to explore road information under a dual view. In the GDVP, we first design the road feature saliency graph reasoning module (RFSG) and road structure homogeneous space graph reasoning module (RSHS) based on the features and structures. The RFSG obtains the similarity of road feature information in different regions as a similarity matrix to participate in the graph reasoning process, ensuring that equal treatment is given to regions with varying saliency levels. The RSHS projects the road structure onto a corresponding homogeneous space while aggregating and interacting with information through graph convolution, thereby enhancing the network’s perception of roads in diverse locations. Second, because multi-view information usually shares a potential common representation, a multi-view information fusion and alignment strategy (MVFA) is designed to comprehensively model road information. Experimental results obtained from two public datasets indicate that RDPGNet outperforms other state-of-the-art networks.

In optical remote sensing images,roads are easily affected by multiple factors such as obstruc-tions,pavement materials,and surrounding environments,resulting in blurred features.However,even if existing road extraction methods enhance their feature perception capabilities,they still suffer from a large number of misjudgments in feature-blurred areas.To address the above issues,this paper proposed the road extraction network based on GCN guided model viewpoint(RGGVNet).RGGVNet adopted the en-coder-decoder structure and designed a GCN based viewpoint guidance module(GVPG)to repeatedly guide the model viewpoint at the connection of the encoder and decoder,thereby enhancing attention to fea-ture blurred areas.GVPG took advantage of the fact that the GCN information propagation process had the characteristic of average feature weight,used the road salience levels in different areas as a Laplacian matrix,and participated in GCN information propagation to realize the guidance model perspective.At the same time,a dense guidance viewpoint strategy(DGVS)was proposed,which uses dense connections to connect the encoder,GVPG module,and decoder to each other to ensure effective guidance of model viewpoints while alleviating optimization difficulties.In the decoding stage,a multi-resolution feature fu-sion module(MRFF)was designed to minimize the information offset and loss of road features of different scales in the feature fusion and upsampling process.In two public remote sensing road datasets,the IoU of our method reached 65.84%and 69.36%,respectively,and the F1-score reached 79.40%and 81.90%,respectively.It can be seen from the quantitative and qualitative experimental results that the performance of our method is superior to other mainstream methods.

A target area guided RRT* robotic arm path planning algorithm (TA-RRT*) is proposed to address the issues of low planning efficiency, poor path quality, and improper robotic arm pose in the traditional RRT* algorithm for robotic arm path planning. Firstly, with the traditional RRT* algorithm as the foundation, a target bias strategy is introduced and a spherical subset constraint sampling is utilized to narrow the sampling range and guide the expansion of the new node towards the target point, enhancing target orientation. By employing a direct connection strategy for new nodes, the algorithm is enabled to converge faster and the speed of path generation is improved. Secondly, by removing redundant points from the initial planning path and transforming it into a smooth path using a cubic B-spline curve, the quality of the path is improved. Finally, the position of the robotic arm is constrained. The reachability of the robotic arm linkage pose is ultimately determined through the inverse kinematics of the robotic arm, and the envelope box model is used to determine whether the robotic arm is collided with obstacles. Experimental results show that the TA-RRT* algorithm outperforms the RRT* algorithm in terms of sampling frequency, planning time, path length, and smoothness in 2D and 3D scenes, verifying the correctness and feasibility of this method. Both the robotic arm simulation experiments and the test results in real environment demonstrate that when adding pose constraints to the planned trajectory of the robotic arm during operation, the joints of the robotic arm do not collide with obstacles during the execution of the planned paths and exhibit good stability.

The extraction of buildings in aerial remote sensing applications is an important and challenging task. Most existing methods extract buildings based on local area attention, ignoring the loss of accuracy due to the global structure of the building. However, global structural features of buildings with strong coupling relationships in complex scenes are difficult to extract, such as the edges and bodies of buildings, leading to discontinuous results. Therefore, multiscale decoupled body and edge supervision network (MDBES-Net), which can consider both edge optimization and inner consistency, is proposed to solve these problems. MDBES-net consists of the body-mask-edge consistency constraint base network (BMECC), decoupling the body and edge aware module (DBEA), and the channel decoupled attention module (CDA). First, body-mask-edge consistency constraint supervision is established by body and edge labels to jointly improve the segmentation effect in the BMECC base network. Second, in the mutiscale DBEA module, building features are warped by a learnable flow field to make body parts more consistent and edges more detailed. Finally, the CDA module performs adaptive calibration of the recoupled feature map channel response to minimize external background noise interference. Experiments on the open Massachusetts building dataset, WHU Building Dataset show that the proposed MDBES-Net can accurately extract buildings in complex scenarios, enabling complete building segmentation with refined boundaries and improved internal consistency.

A model was proposed to address issues with low segmentation accuracy,leakage of tiny cracks,and background interference in the segmentation process of concrete surface cracks.The model combined linear guidance and mesh optimization for crack segmentation.Firstly,the backbone network was enriched with a multi-branch linear guidance module.The network's ability to represent the linear structure of cracks was boosted by adaptive single-dimensional pooling.This facilitated the establishment of connections between cracks in different areas,enhanced the capability to perceive global context data,and improved the network's segmentation accuracy.Then,a module for mesh detail optimization is pro-posed,which divides the entire spatial domain into several spatial meshes through the three steps of parti-tioning,optimization,and merging.The fine cracks' information in the spatial meshes was extracted to prevent the leakage of fine cracks.Finally,a mixed attention module was embedded in the skip connec-tions of the backbone network,highlighting crack features in the two-dimensional space and channels while also reducing background interference.On the Deepcrack537,Crack500,and CFD crack datasets,the proposed model achieves IoU values of 77.07%,58.96%,and 56.55%,respectively.The F1-score val-ues also performs well,achieving 87.05%,74.19%,and 72.24%,respectively.These results are signif-icantly better than those of most existing methods,with superior segmentation accuracy.

Crack detection is an important part of building structural health monitoring. However, the traditional convolution is difficult to capture the characteristics of tiny concrete cracks and the complex topology, which ultimately leads to the misclassification of crack segmentation. To address the above problems, a concrete crack detection network based on local topology and global group attention (LTGGNet) is proposed. In the encoding part, a deformable local attention module (DLA) is designed to improve the ability of the network to extract tiny crack features and complex topologies. It uses deformable convolution to extract the crack regional features and topology information, and strengthens the network’s discriminability of the tiny crack pixel features. In the decoding part, a global grouping context aggregation (GGC) module is proposed to enhance the global modeling ability of the network to capture images of crack structures. It uses the self-attention mechanism to establish the correlation representation of the global information and enhance the long-range dependency of the topology, thereby achieving the accurate localization of cracks and the recovery of details. The experimental results show that the proposed network achieves 75.84%, 86.26%, 59.20%, and 74.37% in the metrics IoU and F1-Score, respectively, on the public crack datasets Deepcrack and Crack500. Compared to the comparison network, the proposed network has a better concrete crack segmentation effect. The ablation experiments further verify the effectiveness of the network.

针对复杂构件装配生产过程中虚实交互困难、装配效率低等问题,提出一种面向复杂构件装配的数字孪生模型建模方法.首先设计了复杂构件装配数字孪生系统架构,然后从几何模型、工艺模型、行为逻辑模型和约束规则模型4个维度构建了复杂构件装配数字孪生虚拟模型,实现了对复杂构件装配物理实体全面真实地刻画与描述.设计并开发了复杂构件装配数字孪生系统,以鲁班锁装配为实例验证了建模方法的有效性,为复杂构件的装配提供了解决方案.

Thermal discomfort body language has been shown to be a psychological representation of personnel's particular thermal comfort. Individual thermal comfort differences are ignored in public building settings with random personnel flow. To solve this issue, we suggested a Bayesian group thermal dissatisfaction rate prediction model based on thermal discomfort body language expression and subsequently implemented intelligent indoor temperature and humidity control. The PMV-PPD model was utilized to represent the group's overall thermal comfort and to create a prior distribution of thermal dissatisfaction rate. To acquire the dynamic distribution of temperature discomfort body language, data on thermal discomfort body language expression were collected in a real-world office setting experiment. Based on Bayesian theory, we used personalized thermal discomfort body language expressions to modify the group's universal thermal comfort and realized the assessment of the thermal dissatisfaction rate by combining commonality and personalization. Finally, a deep reinforcement learning system was employed to achieve intelligent indoor temperature and humidity control. The results show that when commonality and personalized thermal comfort differences are combined, real-time prediction of thermal dissatisfaction rate has high prediction accuracy and good model performance, and the prediction model provides a reference basis for reasonable indoor temperature and humidity settings.

A two- stage dynamic multi-object positioning and grasping method is proposed to solve the problem of fast and accurate grasping of various types of dynamic objects on a factory assembly line. In the first stage, the proposed multiscale context-aware single- branch fusion semantic segmentation network is used to obtain the mask area of the target object: first, the feature extraction network adopts a single- branch structure, which reduces the number of network parameters while ensuring the extraction of rich spatial information and high- level semantic information; subsequently, the feature fusion network improves the expression ability of spatial data and semantic information through the bilateral guided feature fusion module; finally, the feature enhancement network is designed, and the feature assisted convergence module is embedded in the shallow and deep networks to accelerate the convergence speed of the network. In the second stage, a quick pose estimation strategy based on contour point detection is applied to predict the optimum posture of the grasping point in the mask region. The test results on the self- built dataset and the pipeline platform grab experiments demonstrate that the proposed method can detect and predict the position and posture of the object grab points in real time and accurately complete the object grab. Furthermore, its segmentation accuracy, prediction time, and grab success rate are better than the comparison method.

Deep learning methods have made significant progress in the field of hyperspectral image (HSI) classification. However, these methods often rely on a large number of labeled samples, parameters, and computational resources to achieve state-of-the-art performance, which limits their applicability. To address these issues, this paper proposes a lightweight multiview mask contrastive network (LMCN) for HSI classification under small-sample conditions. Considering the influence of irrelevant bands, we construct two views in an HSI scene using band selection and principal component analysis (PCA). To enhance instance discriminability, we propose a combination of self-supervised mask learning and contrastive learning in the design of LMCN. Specifically, we train corresponding masked autoencoders using the obtained views and utilize the feature extraction part of the autoencoder as an augmentation function, conducting unsupervised training through contrastive learning. To reduce the number of parameters, we employ lightweight Transformer modules to construct the autoencoder. Experimental results demonstrate the superiority of this approach over several advanced supervised learning methods and few-shot learning methods under small-sample conditions. Furthermore, this method exhibits lower computational costs. Our code is available at https://github.com/Winkness/LMCN.git .

受公共建筑人员负荷动态变化带来的热环境非均匀分布及建筑热惰性、室内热质扩散等因素的影响,空调控制系统的调节动作难以跟随室内负荷的区域性动态变化,室温响应滞后于空调系统的调节动作,导致环境品质调控不佳、能源浪费.针对该问题,提出了人员负荷视觉动态估计的热不均匀空间温度分区补偿控制方法.首先,设计了室内人员计数与定位模型,采用计算机视觉方法实时获取人员负荷及其空间分布.其次,设计了分区主从补偿控制策略,从控制模块以主控制模块输出为输入,在调节建筑空间总风量的基础上采用开环控制调节各区域风量,改善了室温响应滞后情况;主控制模块由多个区域温度控制回路构成,根据人员负荷变化估计室内温度变化趋势,通过区域补偿运算提高了所需供冷量的预测能力.最后,对小型办公建筑和大型活动中心建筑进行了仿真实验,结果表明,本文调控方法在降低能耗的同时能保持舒适的区域环境,并缩短系统的响应时间.