期刊：
Postharvest Biology and Technology,2026年231:113935 ISSN：0925-5214
通讯作者：
Shan Zeng
作者机构：
[Yixiao Wang; Wei Tao; Shan Zeng; Si Xia; Bing Li; Chaoxian Liu] School of Mathematics and Computer Science, Wuhan Polytechnic University, Wuhan 430023, China
通讯机构：
[Shan Zeng] S;School of Mathematics and Computer Science, Wuhan Polytechnic University, Wuhan 430023, China
摘要：
Hyperspectral imaging technology has been widely used in the non-destructive testing of fruits. However, the curvature change on the surface of fruits leads to differences in the incident angle between different regions and a fixed light source, resulting in distortion of the reflectance in the edge regions. This distortion poses a severe challenge to the detection of early bruise. To solve the problem of inaccurate bruise detection caused by spectral distortion, this paper proposes a method for fruit bruise detection by fusing 3D point clouds with hyperspectral data. Specifically, this paper constructs a fusion architecture for synchronously acquiring heterogeneous data from hyperspectral and point cloud depth cameras. The original data are processed using a joint registration algorithm with spatial geometric constraints to establish a mapping model. Building upon the mapping model, this paper further proposes a parameter-adaptive Lambert-based geometrical influence correction method, which adaptively adjusts the correction coefficient for each pixel using an adaptive polynomial kernel function. Pixel-level correction is achieved via the fusion of spatial-spectral features, resulting in improved reflectance spectrum uniformity. Using the corrected hyperspectral data, this paper designs a fruit bruised detection model for early bruise detection. The experimental results show that the proposed method effectively improves the accuracy of bruise detection in the edge region by up to 26 % across multiple apple varieties, thereby offering a new technical idea for the quality detection of geometrically irregular agricultural products.
Hyperspectral imaging technology has been widely used in the non-destructive testing of fruits. However, the curvature change on the surface of fruits leads to differences in the incident angle between different regions and a fixed light source, resulting in distortion of the reflectance in the edge regions. This distortion poses a severe challenge to the detection of early bruise. To solve the problem of inaccurate bruise detection caused by spectral distortion, this paper proposes a method for fruit bruise detection by fusing 3D point clouds with hyperspectral data. Specifically, this paper constructs a fusion architecture for synchronously acquiring heterogeneous data from hyperspectral and point cloud depth cameras. The original data are processed using a joint registration algorithm with spatial geometric constraints to establish a mapping model. Building upon the mapping model, this paper further proposes a parameter-adaptive Lambert-based geometrical influence correction method, which adaptively adjusts the correction coefficient for each pixel using an adaptive polynomial kernel function. Pixel-level correction is achieved via the fusion of spatial-spectral features, resulting in improved reflectance spectrum uniformity. Using the corrected hyperspectral data, this paper designs a fruit bruised detection model for early bruise detection. The experimental results show that the proposed method effectively improves the accuracy of bruise detection in the edge region by up to 26 % across multiple apple varieties, thereby offering a new technical idea for the quality detection of geometrically irregular agricultural products.
摘要：
In the industrial routine inspection process, accurately tracking pipeline defects is essential for ensuring safe operation. However, this accuracy is adversely affected by the complexity of pipeline internal environment, similarity in the appearance of different defects, and difference in the appearance of the same defect with the movement of the camera used. Herein, we present a novel model called global prototype tracking network (GP-Net), which integrates prototype learning, global attention, and global tracking to accurately track the defects. The proposed model mitigates the feature bias caused by prototype singularity via iterative clustering. A dilated attention method is applied at multiple scales to resolve the redundant computations in global attention. Our experiments demonstrated that GP-Net achieved multi-object tracking accuracy, identity F1 score, and multi-object tracking precision of 79.6%, 85.5%, and 93.3%, respectively. Therefore, we demonstrated the effectiveness of GP-Net in tracking pipeline defects, thereby exploiting the potential application of deep learning in industrial maintenance. The source code and partial dataset are publicly available at https://github.com/cui19981127/GP-Net .
In the industrial routine inspection process, accurately tracking pipeline defects is essential for ensuring safe operation. However, this accuracy is adversely affected by the complexity of pipeline internal environment, similarity in the appearance of different defects, and difference in the appearance of the same defect with the movement of the camera used. Herein, we present a novel model called global prototype tracking network (GP-Net), which integrates prototype learning, global attention, and global tracking to accurately track the defects. The proposed model mitigates the feature bias caused by prototype singularity via iterative clustering. A dilated attention method is applied at multiple scales to resolve the redundant computations in global attention. Our experiments demonstrated that GP-Net achieved multi-object tracking accuracy, identity F1 score, and multi-object tracking precision of 79.6%, 85.5%, and 93.3%, respectively. Therefore, we demonstrated the effectiveness of GP-Net in tracking pipeline defects, thereby exploiting the potential application of deep learning in industrial maintenance. The source code and partial dataset are publicly available at https://github.com/cui19981127/GP-Net .
摘要：
To enhance the efficiency of banana quality management, an automated freshness grading system based on real-time image processing was developed. The system integrates a camera for image acquisition and employs portable edge devices (Nvidia Jetson Orin NX or Jetson TX2) to classify banana freshness into four levels, displaying results along with confidence scores in real-time. A lightweight YOLO-based banana freshness detection method (YOLO-BFD), built on the YOLOv7-tiny framework, was designed to enable high-accuracy, real-time processing on edge devices. Key optimizations—including reduced model width factors, the LightSP module, VoVSSCSP, and Ghost convolutions—resulted in a 118.4 % increase in detection speed and a 1.4 % improvement in mean average precision (mAP) over the baseline, along with a 5 % increase in precision, 1.5 % increase in recall, and a 3 % increase in F1 score. Additionally, the system achieved a 93.5 % reduction in computational costs and a 41.9 % decrease in model parameters. The system demonstrated an average accuracy of 97.8 % in freshness grading, with real-time detection speeds of 61 frames per second on the Jetson Orin NX and 29 frames per second on the Jetson TX2. Overall, the proposed system provides a practical and efficient solution for portable banana quality assessment in agricultural applications.
To enhance the efficiency of banana quality management, an automated freshness grading system based on real-time image processing was developed. The system integrates a camera for image acquisition and employs portable edge devices (Nvidia Jetson Orin NX or Jetson TX2) to classify banana freshness into four levels, displaying results along with confidence scores in real-time. A lightweight YOLO-based banana freshness detection method (YOLO-BFD), built on the YOLOv7-tiny framework, was designed to enable high-accuracy, real-time processing on edge devices. Key optimizations—including reduced model width factors, the LightSP module, VoVSSCSP, and Ghost convolutions—resulted in a 118.4 % increase in detection speed and a 1.4 % improvement in mean average precision (mAP) over the baseline, along with a 5 % increase in precision, 1.5 % increase in recall, and a 3 % increase in F1 score. Additionally, the system achieved a 93.5 % reduction in computational costs and a 41.9 % decrease in model parameters. The system demonstrated an average accuracy of 97.8 % in freshness grading, with real-time detection speeds of 61 frames per second on the Jetson Orin NX and 29 frames per second on the Jetson TX2. Overall, the proposed system provides a practical and efficient solution for portable banana quality assessment in agricultural applications.
作者：
Lv, Site;Wu, Hongan;Zeng, Shan;Yu, Chen;Yang, Ke
期刊：
Mechanical Systems and Signal Processing,2025年234:112826 ISSN：0888-3270
通讯作者：
Shan Zeng
作者机构：
[Lv, Site; Zeng, Shan; Yu, Chen; Yang, Ke] School of Mathematics and Computer Science, Wuhan Polytechnic University, Wuhan, Hubei 430023, China;[Wu, Hongan] Hubei Key Laboratory for Efficient Utilization and Agglomeration of Metallurgic Mineral Resources, Wuhan University of Science and Technology, Wuhan 430081, China
通讯机构：
[Shan Zeng] S;School of Mathematics and Computer Science, Wuhan Polytechnic University, Wuhan, Hubei 430023, China
摘要：
When mechanical equipment fails, the fault characteristics are often interfered by adjacent components. Therefore, how to well characterize the time-varying laws of multi-component signals containing adjacent components has always been a difficulty and research hotspot in the application of time–frequency analysis (TFA) technologies in mechanical fault diagnosis. In this paper, a new TFA method is proposed, called the Multi-scale chirplet synchroextracting transform (MCSET). On the basis of chirplet transform (CT), by using two additional parameters, the window rotation step is divided into two sections within each window length to more accurately match the instantaneous frequency (IF) trajectory of the nonlinear frequency modulation signal. In this way, the energy-concentrated time–frequency (TF) distribution of the multi-component signal containing adjacent components can be obtained. Moreover, to further improve the TF resolution, a new frequency estimation operator is constructed using the idea of synchronous extraction to more accurately capture the IF variation law of multi-component signals and keep the energy highly concentrated. MCSET can well characterize the dynamic characteristics of components adjacent to the IF trajectory, and as a parameterized TFA technique, it maintains good noise robustness. In simulation and experiments, compared with other similar advanced TFA techniques, the results can verify the effectiveness of the proposed method and its superiority in processing complex non-stationary signals with adjacent IF trajectories.
When mechanical equipment fails, the fault characteristics are often interfered by adjacent components. Therefore, how to well characterize the time-varying laws of multi-component signals containing adjacent components has always been a difficulty and research hotspot in the application of time–frequency analysis (TFA) technologies in mechanical fault diagnosis. In this paper, a new TFA method is proposed, called the Multi-scale chirplet synchroextracting transform (MCSET). On the basis of chirplet transform (CT), by using two additional parameters, the window rotation step is divided into two sections within each window length to more accurately match the instantaneous frequency (IF) trajectory of the nonlinear frequency modulation signal. In this way, the energy-concentrated time–frequency (TF) distribution of the multi-component signal containing adjacent components can be obtained. Moreover, to further improve the TF resolution, a new frequency estimation operator is constructed using the idea of synchronous extraction to more accurately capture the IF variation law of multi-component signals and keep the energy highly concentrated. MCSET can well characterize the dynamic characteristics of components adjacent to the IF trajectory, and as a parameterized TFA technique, it maintains good noise robustness. In simulation and experiments, compared with other similar advanced TFA techniques, the results can verify the effectiveness of the proposed method and its superiority in processing complex non-stationary signals with adjacent IF trajectories.
通讯机构：
[Huang, HX ] W;Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan 430048, Peoples R China.
关键词：
To visualize the research hotspots and development direction of green ships in the fields of technology;operation and policy measures;this subsection applies the Co-occurrence Analysis of VOSviewer to generate a visual map. VOSviewer software's clustering algorithm is primarily based on the correlation intensity algorithm;which selects high-frequency keywords in the text for clustering analysis;and reflects the research theme of the discipline through high-frequency keywords. The analysis
摘要：
With the increased global focus on environmental protection, green ships have emerged as the cornerstone to the maritime industry's long-term development. Based on the knowledge graph method, this paper uses bibliometric analysis to examine the research progress of green ships in a multidimensional perspective. Firstly, sort out the definition and connotation of green ships, and clarify their greening requirements in the whole life cycle. Bibliometric analysis of relevant literature using the bibliometrix package; secondly, combined with the virtual analysis method-VOSviewer software, it explores the current research status of green ships from the multi-dimensional perspectives of technical, operational and policy. The results indicate that the hot spots in technical measures are “green energy”, “ship optimization” and “waste heat recovery”; the hot spots in operational measures are “speed optimization” and “liner shipping”; and the hot spots in policy measures are “market mechanism” and “international regulations”.
With the increased global focus on environmental protection, green ships have emerged as the cornerstone to the maritime industry's long-term development. Based on the knowledge graph method, this paper uses bibliometric analysis to examine the research progress of green ships in a multidimensional perspective. Firstly, sort out the definition and connotation of green ships, and clarify their greening requirements in the whole life cycle. Bibliometric analysis of relevant literature using the bibliometrix package; secondly, combined with the virtual analysis method-VOSviewer software, it explores the current research status of green ships from the multi-dimensional perspectives of technical, operational and policy. The results indicate that the hot spots in technical measures are “green energy”, “ship optimization” and “waste heat recovery”; the hot spots in operational measures are “speed optimization” and “liner shipping”; and the hot spots in policy measures are “market mechanism” and “international regulations”.
作者机构：
[Zhao, Jiemei] School of Mathematics and Computer Science, Wuhan Polytechnic University, Wuhan, 430023, China. Electronic address: jiemeizhao@whpu.edu.cn;[Wang, Fenglin] School of Mathematics and Computer Science, Wuhan Polytechnic University, Wuhan, 430023, China
通讯机构：
[Zhao, Jiemei] S;School of Mathematics and Computer Science, Wuhan Polytechnic University, Wuhan, 430023, China. Electronic address:
摘要：
The synchronization control problem of delayed inertial memristive neural networks (DIMNNs) with reaction-diffusion terms is addressed. First, by differential inclusions and a reduced-order approach to analyze the DIMNNs. Second, an adaptive pinning control method is introduced to achieve synchronization of drive-and-response systems. By applying inequality technology and Green’s Formula, a criterion is obtained to ensure the synchronization of DIMNNs with reaction-diffusion terms. Furthermore, the proposed control scheme requires fewer controlled nodes compared to the full-state feedback method, exhibits robustness against parameter uncertainties, and automatically adapts to varying network conditions. Finally, a numerical example is presented to confirm the theoretical results and demonstrate the effectiveness of the proposed approach.
The synchronization control problem of delayed inertial memristive neural networks (DIMNNs) with reaction-diffusion terms is addressed. First, by differential inclusions and a reduced-order approach to analyze the DIMNNs. Second, an adaptive pinning control method is introduced to achieve synchronization of drive-and-response systems. By applying inequality technology and Green’s Formula, a criterion is obtained to ensure the synchronization of DIMNNs with reaction-diffusion terms. Furthermore, the proposed control scheme requires fewer controlled nodes compared to the full-state feedback method, exhibits robustness against parameter uncertainties, and automatically adapts to varying network conditions. Finally, a numerical example is presented to confirm the theoretical results and demonstrate the effectiveness of the proposed approach.
作者机构：
Hubei Key Laboratory of Digital Finance Innovation, Hubei University of Economics, 430205, Wuhan, Hubei, China;School of Information Engineering, Hubei University of Economics, 430205, Wuhan, Hubei, China;[Zhi Yang] College of Electronics and Information Engineering, Sichuan University, 610065, Chengdu, China;[Yanfei Zhu] School of Foreign Languages, Sun Yat-sen University, 510275, Guangzhou, China;[Libin Lu] School of Mathematics and Computer Science, Wuhan Polytechnic University, 430023, Wuhan, China
会议名称：
Brain Informatics: 17th International Conference, BI 2024, Bangkok, Thailand, December 13–15, 2024, Proceedings, Part II
摘要：
Early diagnosis of mild cognitive impairment (MCI) is crucial for the effective treatment and intervention of neurodegenerative diseases. Effective connectivity is one kind of brain network, which is helpful for analyzing the pathogenic mechanism of MCI. It is challenging to model causal relationships between brain regions from multimodal imaging data. This study proposes a new method for brain network causality modeling based on the structure-guided spatiotemporal diffusion model (SSDM), aiming to improve the accuracy of MCI diagnosis. By utilizing the advanced diffusion models, we introduced structural connectivity to guide the transformer-based network to learn topological and spatiotemporal features, which can better remove uncorrelated noise and improve effective connectivity estimation. The proposed model can not only generate temporal features of brain regions with individual differences but also construct discriminable effective connectivities. Experiments on the ADNI dataset demonstrate the effectiveness of our model, showing a certain improvement in diagnostic accuracy compared with competing methods. In addition, by analyzing the effective connectivities, our model predicts abnormal brain connections that are highly correlated with MCI. Overall, the framework proposed in this paper provides insights into the potential neurobiological mechanisms of MCI, which may promote early intervention strategies.
期刊：
International Journal of Geographical Information Science,2025年 ISSN：1365-8816
通讯作者：
Yao, Y
作者机构：
[Tang, Mi; Yu, Chenglong; Yao, Y; Yao, Yao; Hu, Zhihui; Zhu, Geyuan; Guan, Qingfeng; Zhang, Xiang; Ma, Yueheng] China Univ Geosci, Sch Geog & Informat Engn, Wuhan, Hubei, Peoples R China.;[Yu, Chenglong; Yao, Y; Yao, Yao; Zhu, Geyuan; Zhang, Xiang] LocationMind Inc, LocationMind Inst, Chiyoda, Tokyo, Japan.;[Hu, Huanjun] Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan, Peoples R China.;[Yao, Y; Yao, Yao; Guan, Qingfeng] China Univ Geosci, Natl Engn Res Ctr Geog Informat Syst, Wuhan, Hubei, Peoples R China.;[Yao, Y; Yao, Yao] Hitotsubashi Univ, Hitotsubashi Inst Adv Study, Kunitachi, Tokyo, Japan.
通讯机构：
[Yao, Y ] C;China Univ Geosci, Sch Geog & Informat Engn, Wuhan, Hubei, Peoples R China.;LocationMind Inc, LocationMind Inst, Chiyoda, Tokyo, Japan.;China Univ Geosci, Natl Engn Res Ctr Geog Informat Syst, Wuhan, Hubei, Peoples R China.;Hitotsubashi Univ, Hitotsubashi Inst Adv Study, Kunitachi, Tokyo, Japan.
关键词：
Land-use classification;LandGPT;LLM tuning;prompt engineering;multi-source geospatial data
摘要：
Actual land parcels vary significantly in size and complexity. Previous studies were limited by existing technical methods for fine-grained land use classification. The emergence of multimodal large language models offers new techniques for image classification, but their application in land use classification remains unexplored. This study presents LandGPT, a multimodal large language model trained on the CN-MSLU-100K dataset, covering fine-grained land use classification of irregular parcels. This study proposes a trans-level discrimination framework to improve LandGPT’s ability to classify fine-grained land use. Under this framework, LandGPT achieves a discrimination accuracy of 89.7% and a Kappa coefficient of 0.85 for fine-grained land use categories, outperforming state-of-the-art models with a 48.33% accuracy improvement. In some challenging categories, the improvement reaches nearly 1500%. This study finds that training with multi-source remote sensing image data improved LandGPT’s accuracy by 15.79% compared to single-image data. This study explores Prompt engineering based on LandGPT. The optimal prompt paradigm offers fine-grained categories and guides the model for accurate classification, reducing errors from LLM hallucinations. This study pioneeringly explores the application of large language models in the land use domain and offers a new solution for fine-grained land use classification.
摘要：
The phenotypic characteristics of pomegranates at different growth stages significantly influence yield, making real-time detection essential for optimizing production. Additionally, complex environmental conditions complicate this process. Current research on pomegranate growth stage detection is scarce, with many studies ignoring environmental interference. To address these gaps, this study proposes a hybrid and efficient scale-aware pomegranate detection transformer model (HESP-DETR) based on Real-Time DEtection TRansformer (RT-DETR). Firstly, reparameterized cheap-operation layer aggregation network (RCLAN) was proposed to aggregate multi-layer semantic information, thereby reducing redundant feature processing and optimizing inference efficiency. Secondly, diverse branch block fusion module (DBFusion) was proposed to merge local and global semantics through multiple branch paths, preserving information in complex scenes while alleviating the computational burden. Finally, multi-scale multi-head self-attention (MMSA) was improved and introduced into the transformer encoder to enhance multi-scale feature processing and strengthen the interaction between local and global features, thereby boosting the model’s global perception ability. Experimental results show that HESP-DETR achieves a mean Average Precision (mAP 50 ) of 93.4%, outperforming RT-DETR by increasing mAP 50 by 1% and reducing parameter count, Giga Floating-point Operations Per Second (GFLOPs), weight size and inference time by 37.5%, 36.5%, 37% and 34% respectively. Robustness experiments further confirm that HESP-DETR demonstrates superior anti-interference capability compared to other mainstream models. Overall, HESP-DETR achieves a favorable balance between accuracy, model size, and speed, delivering a positive impact on the practical detection tasks of pomegranate growth stages.
The phenotypic characteristics of pomegranates at different growth stages significantly influence yield, making real-time detection essential for optimizing production. Additionally, complex environmental conditions complicate this process. Current research on pomegranate growth stage detection is scarce, with many studies ignoring environmental interference. To address these gaps, this study proposes a hybrid and efficient scale-aware pomegranate detection transformer model (HESP-DETR) based on Real-Time DEtection TRansformer (RT-DETR). Firstly, reparameterized cheap-operation layer aggregation network (RCLAN) was proposed to aggregate multi-layer semantic information, thereby reducing redundant feature processing and optimizing inference efficiency. Secondly, diverse branch block fusion module (DBFusion) was proposed to merge local and global semantics through multiple branch paths, preserving information in complex scenes while alleviating the computational burden. Finally, multi-scale multi-head self-attention (MMSA) was improved and introduced into the transformer encoder to enhance multi-scale feature processing and strengthen the interaction between local and global features, thereby boosting the model’s global perception ability. Experimental results show that HESP-DETR achieves a mean Average Precision (mAP 50 ) of 93.4%, outperforming RT-DETR by increasing mAP 50 by 1% and reducing parameter count, Giga Floating-point Operations Per Second (GFLOPs), weight size and inference time by 37.5%, 36.5%, 37% and 34% respectively. Robustness experiments further confirm that HESP-DETR demonstrates superior anti-interference capability compared to other mainstream models. Overall, HESP-DETR achieves a favorable balance between accuracy, model size, and speed, delivering a positive impact on the practical detection tasks of pomegranate growth stages.
摘要：
OBJECTIVES: To analyse prescribing patterns of cyclooxygenase-2 inhibitor for low-back pain patients with cardiovascular comorbidities in American outpatient settings. METHODS: The data of this retrospective, cross-sectional study were from the 2007-2019 National Ambulatory Medical Care Survey except 2017 for which data were not available. Data related to low-back pain patients of either gender aged ≥20 years. Those having cardiovascular comorbidities were placed in group A, while those without such comorbidities were placed in group B. Descriptive statistics were employed to evaluate visit characteristics, stratified by cyclooxygenase-2 inhibitor use. Multivariable logistic regression analysis was utilised to assess factors associated with cyclooxygenase-2 inhibitor prescriptions. Data was analysed using R 4.1.2. RESULTS: Of the 242.65 million patients with 107.19(44.2%) females, 76.83 million (31.7%) were in group A and 165.82 million (68.3%) in group B. Compared to group B patients, those in group A were older (62.0±14.1years vs 49.7±16.1 years, p<0.01) and had a higher prevalence of cyclooxygenase-2 inhibitor use (p=0.01). Overall, 5.2 million (2.14%) patients were prescribed cyclooxygenase-2 inhibitors. Those using cyclooxygenase-2 inhibitor exhibited a higher prevalence of cardiovascular comorbidities (p=0.01), especially hypertension (p=0.01), and were older in age (p<0.01). Older age (odds ratio = 1.019, 95% confidence interval: 1.003-1.035; p<0.05) and higher prevalence of cardiovascular comorbidities (odds ratio = 1.638, 95% CI: 1.017-2.637; p<0.05) were associated with increased likelihood of receiving cyclooxygenase-2 inhibitor prescriptions. CONCLUSIONS: Cyclooxygenase-2 inhibitor use was positively correlated with age and cardiovascular comorbidities among low-back pain patients in American ambulatory care, suggesting potential contradiction to current medication guidelines and heightened risk of adverse cardiovascular events.
摘要：
The corrosion of steel reinforcement in concrete structures due to chloride exposure poses significant financial and environmental risks. An accurate assessment of chloride diffusion is essential for predicting the service life of steel-reinforced concrete. This article develops six models for estimating the chloride diffusion coefficient (CDC) in concrete, considering various exposure conditions like tidal, splash, atmospheric, and submerged environments. The models utilize a hybrid approach combining the adaptive neuro-fuzzy inference system (ANFIS) and least square support vector regression (LSSVR), enhanced by an improved arithmetic optimization algorithm (IAA). The IAA merges the arithmetic optimization algorithm (AOA) with the Aquila optimizer (AO) to address AOA's limitations. The models undergo sensitivity analysis using the Fourier Amplitude Sensitivity Test (FAST), revealing that the curing mechanism (CM) is the most influential factor, with a sensitivity value of 0.993. The study demonstrates that LSSVR and ANFIS-based methods are highly effective in predicting CDC. Among them, the ANFIS combined with IAA outperforms others regarding reliability and accuracy, making it a superior choice for CDC estimation. This robust model could lead to better management and longevity of concrete structures exposed to chloride, mitigating potential risks. The main application of this research is to enhance the durability management and service life prediction of steel-reinforced concrete structures exposed to chloride-laden environments. Engineers and infrastructure managers can better assess the risk and timing of corrosion onset by accurately estimating the CDC, which is crucial for preventive maintenance and cost-effective design.
作者机构：
["Chen, Songnan] School of Mathematics and Computer Science, Wuhan Polytechnic University, Wuhan, Hubei, China;["Chen, Songnan] Foshan Zhongke Innovation Research Institute of Intelligent Agriculture and Robotics, Foshan, Guangdong, China;[Feng, Zhaoxu"] China United Network Communications Co., Ltd. Henan Branch, Zhengzhou, Henan, China. iefengzhaoxu@163.com
摘要：
Recent data-driven deep learning methods for image reflection removal have made impressive progress, promoting the quality of photo capturing and scene understanding. Due to the massive consumption of computational complexity and memory usage, the performance of these methods degrades significantly while dealing with high-resolution images. Besides, most existing methods for reflection removal can only remove reflection patterns by downsampling the input image into a much lower resolution, resulting in the loss of plentiful information. In this paper, we propose a novel transformer-based framework for high-resolution image reflection removal, termed as the Laplacian pyramid-based component-aware transformer (LapCAT). LapCAT leverages a Laplacian pyramid network to remove high-frequency reflection patterns and reconstruct the high-resolution background image guided by the clean low-frequency background components. Guided by the reflection mask through pixel-wise contrastive learning, LapCAT designs a component-separable transformer block (CSTB) which removes reflection patterns from the background constituents through a reflection-aware multi-head self-attention mechanism. Extensive experiments on several benchmark datasets for reflection removal demonstrate the superiority of our LapCAT, especially the excellent performance and high efficiency in removing reflection from high-resolution images than state-of-the-art methods.
摘要：
In the event of disaster-related disruptions to public networks, an air-ground communication-navigation-sensing (CNS) ad-hoc network (ANET) can effectively support emergency response operations by ensuring communication, navigation and sensing capabilities. However, during extreme events such as typhoons and torrential rain, flight operations become impossible for aircraft. Therefore, a critical issue that requires immediate investigation is how to effectively provide communication, navigation, and sensing assistance to emergency regions using only ground-based CNS nodes. This paper presents a communication-navigation-sensing spatiotemporal cooperative optimization algorithm based on ResNet-DDPG. It aims to achieve spatiotemporal cooperative optimization of ground CNS nodes in the absence of aerial relay ad hoc network nodes (RANET nodes). This work has three main components: (1) a neural network-based geospatial fitting model for ANET communication, (2) a ResNet based spatiotemporal feature and CNS node fusion encoding method for state space construction for reinforcement learning, and (3) a deep reinforcement learning algorithm called ResNet-DDPG for deploying ground ANET nodes dynamically. Experimental results indicate that, compared to similar approaches, this method significantly improves communication rate optimization and deployment site effectiveness. This study supports emergency response by utilizing ground ANET, which is crucial for disaster emergency response, command scheduling, and life rescue in extreme disaster scenarios.
作者机构：
[Zhou, Xiao-Wu] Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan 430048, Peoples R China.;[Liu, Qi] Hubei Minzu Univ, Sch Math & Stat, Enshi 445000, Peoples R China.
通讯机构：
[Zhou, XW ] W;Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan 430048, Peoples R China.
关键词：
Direct sum;(L, M)-convex space;L-convex space;M-fuzzifying spaces;separability;additivit.y
摘要：
In this paper, the notion of direct sum of a family of convex spaces is generalized to that of a family of (L, M)-fuzzy convex spaces. Firstly, the related properties between the direct sum of a family of (L, M)-convex spaces and its factor spaces are discussed. Secondly, the (L, M)-fuzzy direct sum convex spaces are characterized by means of its some level L-direct sum convex spaces. Finally, the additivity of separability (S-1, sub-S-0, S-0, S-1, S-2) are investigated.
摘要：
We study the Schrödinger-Poisson-Slater equation $ \begin{equation*} \left\{\begin{array}{lll} -\Delta u + \lambda u + \big(|x|^{-1} \ast |u|^{2}\big)u = V(x) u^{ p_ \varepsilon-1 } , \, \text{ in } {\mathbb{R}}^{3}, \\ \int_{{\mathbb{R}}^3}u^2 \, \mathrm{d}x = a, \, \, u > 0, u \in H^{1}(\mathbb{R}^{3}), \end{array} \right. \end{equation*} $ where $ \lambda $ is a Lagrange multiplier, $ V(x) $ is a real-valued potential, $ a\in {\mathbb{R}}_{+} $ is a constant, $ p_{ \varepsilon} = \frac{10}{3} \pm \varepsilon $ and $ \varepsilon>0 $ is a small parameter. In this paper, we prove that it is the positive critical value of the potential $ V $ that affects the existence of single-peak solutions for this problem. Furthermore, we prove the local uniqueness of the solutions we construct.
通讯机构：
[Yang, H ] W;Wuhan Polytech Univ, Coll Math & Comp Sci, Wuhan 430023, Peoples R China.
关键词：
BP neural network;hybrid optimization model;metaheuristic algorithm;renewable energy integration;wind power prediction
摘要：
The integration of intermittent wind power into modern grids necessitates highly accurate forecasting models to ensure stability and efficiency. To address the limitations of traditional backpropagation (BP) neural networks, such as slow convergence and susceptibility to local optima, this study proposes a novel hybrid framework: the Multi-Strategy Coati Optimization Algorithm (SZCOA)-optimized BP neural network (SZCOA-BP). The SZCOA integrates three innovative strategies-a population position update mechanism for global exploration, an olfactory tracing strategy to evade local optima, and a soft frost search strategy for refined exploitation-to enhance the optimization efficiency and robustness of BP networks. Evaluated on the CEC2017 benchmark, the SZCOA outperformed state-of-the-art algorithms, including ICOA, DBO, and PSO, achieving superior convergence speed and solution accuracy. Applied to a real-world wind power dataset (912 samples from Alibaba Cloud Tianchi), the SZCOA-BP model attained an R² of 94.437% and reduced the MAE to 10.948, significantly surpassing the standard BP model (R²: 81.167%, MAE: 18.891). Comparative analyses with COA-BP, BWO-BP, and other hybrid models further validated its dominance in prediction accuracy and stability. The proposed framework not only advances wind power forecasting but also offers a scalable solution for optimizing complex renewable energy systems, supporting global efforts toward sustainable energy transitions.
通讯机构：
[Cheng, C ] H;Hubei Meteorol Serv Ctr, Wuhan 430074, Peoples R China.
关键词：
river flow forecasting;hourly-scale prediction;small catchments;TAM-xLSTM;Qiandongnan region
摘要：
Accurate river flow forecasting is vital for flood warning and water resource management, yet hourly-scale prediction in small catchments remains underexplored despite its importance for rapid response flood control. To address this gap, this study proposes an enhanced temporal attention module xLSTM (TAM-xLSTM) model that combines temporal feature extraction with timestep-level attention to better capture dynamic variations and dependencies. Case studies in the Qiandongnan region demonstrate that TAM-xLSTM substantially outperforms baseline models during wet season forecasting at Panghai Station, reducing RMSE by 9.6%, MAE by 24.1%, and Theil’s U by 6.6%, while increasing NSE by 4.8%. These results highlight the model’s ability to improve short-term river flow prediction in complex mountainous terrain and its potential to support effective flood warning and water resource management.
摘要：
The accurate detection of road defects is crucial for enhancing the safety and efficiency of road maintenance. This study focuses on six common types of pavement defects: transverse cracks, longitudinal cracks, alligator cracking, oblique cracks, potholes, and repair marks. In real-world scenarios, key challenges include effectively distinguishing between the foreground and background, as well as accurately identifying small-sized (e.g., fine cracks, dense alligator cracking, and clustered potholes) and overlapping defects (e.g., intersecting cracks or clustered damage areas where multiple defects appear close together). To address these issues, this paper proposes a Pavement-DETR model based on the Real-Time Detection Transformer (RT-DETR), aiming to optimize the overall accuracy of defect detection. To achieve this goal, three main improvements are proposed: (1) the introduction of the Channel-Spatial Shuffle (CSS) attention mechanism in the third (S3) and fourth (S4) stages of the ResNet backbone, which correspond to mid-level and high-level feature layers, enabling the model to focus more precisely on road defect features; (2) the adoption of the Conv3XC structure for feature fusion enhances the model's ability to differentiate between the foreground and background, which is achieved through multi-level convolutions, channel expansion, and skip connections, which also contribute to improved gradient flow and training stability; (3) the proposal of a loss function combining Powerful-IoU v2 (PIoU v2) and Normalized Wasserstein Distance (NWD) weighted averaging, where PIoU v2 focuses on optimizing overlapping regions, and NWD targets small object optimization. The combined loss function enables comprehensive optimization of the bounding boxes, improving the model's accuracy and convergence speed. Experimental results show that on the UAV-PDD2023 dataset, Pavement-DETR improves the mean average precision (mAP) by 7.7% at IoU = 0.5, increases mAP by 8.9% at IoU = 0.5-0.95, and improves F1 Score by 7%. These results demonstrate that Pavement-DETR exhibits better performance in road defect detection, making it highly significant for road maintenance work.
摘要：
Blended vegetable oil is considered to be a valuable product in the market owing to favourable taste and nutritional composition. The quantification of its contents has notable implications for protecting food safety and consumer interests. Thus, a rapid and non-destructive method is needed to analyse the composition of blended oil. This study established an analytical method combining Raman spectroscopy and prediction models to determine the content of olive oil in a mixture. Competitive adaptive reweighted sampling was employed to select feature bands attributed to β-carotene and unsaturated fatty acids. Various models were used to calculate the mixture proportion, and the importance of characteristic peak intensity affecting the prediction was evaluated via grey relational analysis. The random forest model exhibited superior performance in quantitative analysis, with RMSE and R 2 of 0.0447 and 0.9799, respectively. Overall, this approach was proven to effectively identify blended olive oils, exemplifying its potential in food authentication.
Blended vegetable oil is considered to be a valuable product in the market owing to favourable taste and nutritional composition. The quantification of its contents has notable implications for protecting food safety and consumer interests. Thus, a rapid and non-destructive method is needed to analyse the composition of blended oil. This study established an analytical method combining Raman spectroscopy and prediction models to determine the content of olive oil in a mixture. Competitive adaptive reweighted sampling was employed to select feature bands attributed to β-carotene and unsaturated fatty acids. Various models were used to calculate the mixture proportion, and the importance of characteristic peak intensity affecting the prediction was evaluated via grey relational analysis. The random forest model exhibited superior performance in quantitative analysis, with RMSE and R 2 of 0.0447 and 0.9799, respectively. Overall, this approach was proven to effectively identify blended olive oils, exemplifying its potential in food authentication.
作者机构：
[Huang, Chuan] Wuhan Polytech Univ, Sch Math & Comp Sci, Wuhan 430024, Peoples R China.;[Yang, Yinke; Yang, Zhizhou; Zhang, Jincheng] Changan Univ, Sch Water & Environm, Xian 710064, Peoples R China.;[Yang, Yinke; Yang, Zhizhou; Zhang, Jincheng] Changan Univ, Key Lab Subsurface Hydrol & Ecol Effect Arid Reg, Minist Educ, Xian 710064, Peoples R China.;[Yang, Yinke; Yang, Zhizhou; Zhang, Jincheng] Changan Univ, Key Lab Ecohydrol & Water Secur Arid & Semiarid Re, Minist Water resources, Xian 710064, Peoples R China.;[Yang, Xuyang; Yang, XY] Weinan Normal Univ, Coll Environm & Life Sci, Weinan 714099, Peoples R China.
通讯机构：
[Yang, XY ] W;Weinan Normal Univ, Coll Environm & Life Sci, Weinan 714099, Peoples R China.;Weinan Normal Univ, Key Lab Ecol & Environm River Wetlands Shaanxi Pro, Weinan 714099, Peoples R China.
关键词：
runoff;climate change;human activities;attribution analysis;Dawen River Basin
摘要：
Surface runoff change is significantly influenced by both human activities and climate change. Decoupling their respective contributions to runoff change represents a critical frontier in hydrological research and a pressing challenge for water resource management. This study focuses on the Dawen River Basin, a strategic area for ecological conservation and high-quality development in the lower Yellow River region. By integrating three methodological approaches—empirical models (Precipitation–Runoff Double Mass Curve), conceptual models (elasticity coefficient methods), and hydrological models (Soil and Water Assessment Tool, SWAT)—we systematically quantify the impacts of climate change and human activities on runoff change. A correlation analysis was first applied to screen independent runoff drivers and basin characteristic factors, followed by a random forest algorithm to rank their relative importance. This process informed the establishment of a comprehensive framework for runoff attribution analysis. Results demonstrate that hydrological modeling (SWAT) is the most appropriate method for the Dawen River Basin, revealing human activities as the dominant driver of runoff changes, accounting for 70% to 82%. These findings provide critical insights for guiding sustainable water resource planning and management in anthropogenically stressed basins under a changing environment.
