Applied Soft Computing最新文献

{"title":"Towards efficient compression and communication for prototype-based decentralized learning","authors":"Pablo Fernández-Piñeiro ,&nbsp;Manuel Fernández-Veiga ,&nbsp;Rebeca P. Díaz-Redondo ,&nbsp;Ana Fernández-Vilas ,&nbsp;Martín González Soto","doi":"10.1016/j.asoc.2025.113270","DOIUrl":"10.1016/j.asoc.2025.113270","url":null,"abstract":"<div><div>In prototype-based federated learning, the exchange of model parameters between clients and the master server is replaced by transmission of prototypes or quantized versions of the data samples to the aggregation server. A fully decentralized deployment of prototype-based learning, without a central aggregator of prototypes, is more robust upon network failures and reacts faster to changes in the statistical distribution of the data, suggesting potential advantages and quick adaptation in dynamic learning tasks, e.g., when the data sources are IoT devices or when data is non-iid. In this paper, we face the challenge of designing an efficient prototype-based decentralized learning network by reducing the overheads in communication and computation. This allows enhancing the scalability of the global system, specially for IoT settings with resource-limited devices. First, we compress the prototype size by applying a clustering algorithm. After that, we filter the prototypes to be disseminate using an information-theoretic measure to share only relevant models or models that provide new knowledge to their neighbors. Then, we define a parallel gossip algorithm to disseminate these models within the learning network. Finally, we define a suitable scheduler able to manage the set of prototypes received to optimize the aggregation phase. In order to validate our proposal we present an analysis of the parallel gossip algorithm regarding the age-of-information (AoI). Our experimental results show the communications load can be substantially reduced without decreasing the convergence rate of the learning algorithm.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"178 ","pages":"Article 113270"},"PeriodicalIF":7.2,"publicationDate":"2025-05-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144168424","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Feature fusion-based hand gesture classification with time-domain descriptors and multi-level deep attention network","authors":"Ömer Faruk Alçin ,&nbsp;Deniz Korkmaz ,&nbsp;Hakan Acikgoz","doi":"10.1016/j.asoc.2025.113375","DOIUrl":"10.1016/j.asoc.2025.113375","url":null,"abstract":"<div><div>In conventional human-robot interaction (HRI), it is difficult to provide adaptability by located systems in the human body. Surface Electromyography (sEMG) signals have the potential to meet adaptability in HRI by directly representing movements, and classifying hand gestures with sEMG can be an effective solution to meet the increasing needs of these applications. In this paper, a hybrid and multi-scale convolutional neural network (CNN) model is proposed to obtain an efficient sEMG-based classification approach of human hand gestures. The proposed method includes an effective feature extraction process, including spectral moments, sparseness, irregularity factor, Teager–Kaiser energy, Shannon entropy, Katz fractal dimension, and Higuchi’s fractal dimension, and waveform length. The obtained features are then converted to RGB images. The designed network is built on multi-scale convolutional blocks with residual learning and convolutional blocks, including the CBAM to improve the network performance by focusing on channel and spatial features. Furthermore, a pyramid non-pooling local block is utilized at the end of the network to learn more powerful features and their correlations. Five comprehensive publicly available datasets are evaluated in the experiments, and the obtained results are compared with the benchmark CNN models and network variations with different attention mechanisms. In the comparative evaluations, the CBAM achieves a classification accuracy between 84.62 % and 97.56 % while other attention mechanism results give accuracy values between 82.88 % and 97.17 %. The experiments show that the proposed method gives more accurate and robust classification performance compared with other variations and benchmark models.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"178 ","pages":"Article 113375"},"PeriodicalIF":7.2,"publicationDate":"2025-05-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144168422","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Fusion channel interaction attention network for water jet detection in firefighting robots","authors":"Juxian Zhao ,&nbsp;Wei Li ,&nbsp;Jinsong Zhu ,&nbsp;Zhigang Gao ,&nbsp;Lu Pan ,&nbsp;Zhongguan Liu","doi":"10.1016/j.asoc.2025.113364","DOIUrl":"10.1016/j.asoc.2025.113364","url":null,"abstract":"<div><div>Firefighting robots play a critical role in fire suppression. Ensuring the water stream precisely hits the target during autonomous fire extinguishing is of paramount importance. By visually detecting the landing point of the water jet, closed-loop control of the extinguishing process can be achieved. However, achieving accurate jet landing point localization in complex environments, such as changes in ambient lighting, jet end divergence, and jet breakup, presents a challenging task. To address this, we propose a novel CIA-YOLOX (Channel Interaction Attention–You Only Look Once) model for the precise identification of water jet landing points in firefighting robots using unmanned aerial vehicle (UAV) visual information. First, the model introduces the Triplet Attention (TA) mechanism to capture feature dependencies across different dimensions, enriching feature information. Second, a module named Coordinate Attention Transformer (CA-Trans) is designed to establish long-range dependencies between directional feature vectors, enabling the extraction of precise positional information critical for accurate impact point prediction. Additionally, a Dual-branch Channel Interactive Attention Fusion (DCIAF) module is proposed to enhance feature representation capabilities by facilitating feature complementation through semantic modeling of channel interactions. Experimental results indicate that the proposed model surpasses current state-of-the-art methods in performance while maintaining low computational costs, confirming its efficacy. This approach enhances the robot's ability to perceive complex environments, providing valuable insights for implementing firefighting actions in real-world scenarios.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"178 ","pages":"Article 113364"},"PeriodicalIF":7.2,"publicationDate":"2025-05-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144168423","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Skin lesion segmentation by fusing local and global features using axial shift and spatial state model","authors":"Guangju Li ,&nbsp;Qinghua Huang ,&nbsp;Wei Wang ,&nbsp;Longzhong Liu","doi":"10.1016/j.asoc.2025.113261","DOIUrl":"10.1016/j.asoc.2025.113261","url":null,"abstract":"<div><div>Skin lesions present challenges to medical diagnosis due to their complex features, including shape variations, uneven color, and blurred boundaries. Currently, models based on convolutional neural networks (CNNs) and Transformers often have too many parameters, making them difficult to deploy in resource-limited medical environments while also struggling to balance local and global features. To address this, this paper proposes a Shift-Mamba structure that effectively captures local features through an axial shift mechanism and fuses global features using Mamba’s spatial state model (SSM). Notably, the new model (SM-UNet) designed based on the Shift-Mamba structure has only 0.02 million (M) parameters, making it one of the lightest models available, much lighter than those based on CNN or Transformer architectures. The SM-UNet model was validated on the ISIC 2017 and ISIC 2018 datasets, achieving IoU and Dice scores of 84.04%, 91.15% and 82.50%, 90.23%, respectively. These results surpass those of existing segmentation models, demonstrating the superiority of SM-UNet in the task of skin lesion segmentation. Code is available at <span><span>https://github.com/guangguangLi/SM-UNet</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"178 ","pages":"Article 113261"},"PeriodicalIF":7.2,"publicationDate":"2025-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144155110","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Short-time photovoltaic power forecasting based on Informer model integrating Attention Mechanism","authors":"Weijie Yu ,&nbsp;Yeming Dai ,&nbsp;Tao Ren ,&nbsp;Mingming Leng","doi":"10.1016/j.asoc.2025.113345","DOIUrl":"10.1016/j.asoc.2025.113345","url":null,"abstract":"<div><div>Precise Photovoltaic Power Generation Forecasting (PVGF) is significant for achieving reliable power supply, optimizing energy scheduling, and responding to changing energy market demand for sustainable development. However, Photovoltaic Power (PV) is vulnerable to changes in solar radiation levels and temperature, then result in electricity generation fluctuations. To further enhance the precision of PVGF, we propose a new short-term PVGF method based on Informer model integrating attention mechanism. Firstly, Locally Weighted Scatterplot Smoothing (LOWESS) is introduced to preprocess data, enhancing the stability of the input data. Secondly, Feature Engineering (FE) is used for feature screening. Thirdly, Informer model is improved, termed as Attention-Informer-Attention (AT-Informer-AT) model. Specifically, Attention mechanism (AM) layer is added to the encoder and decoder of Informer model respectively, allowing the model to flexibly adjust the attention to different time series data and effectively capture important patterns in the PV data, thereby enhancing prediction performance and generalization ability. Eventually, the novel prediction approach’s efficiency is confirmed through analyzing the cases of two different power stations in DKASC area, Alice Springs, Australia and Xuhui District, Shanghai, China. The Experimental results demonstrate that the proposed method superiors other models, with the best prediction accuracy and generalization ability.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"178 ","pages":"Article 113345"},"PeriodicalIF":7.2,"publicationDate":"2025-05-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144155113","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Extracting hierarchical relationships of aspects from reviews","authors":"Jiangtao Qiu ,&nbsp;Ling Lin ,&nbsp;Siyu Wang","doi":"10.1016/j.asoc.2025.113335","DOIUrl":"10.1016/j.asoc.2025.113335","url":null,"abstract":"<div><div>Aspect Based Sentiment Analysis (ABSA) attracts significant attention in recent years. Three elements of ABSA including aspect term, aspect, and domain (or entity) present a hierarchical relationships in e-commerce reviews. Extracting the hierarchical relationships can significantly enhance various applications, such as creating user profiles, identifying hierarchical topics, and visualizing review data. In this study, we proposed a framework to tackle this task, consisting of two primary components: a text adversarial autoencoder that efficiently encodes review content, and a deep network that extracts the clusters of aspect terms from review dataset and organizes them to a hierarchical structure using the Student-Teacher paradigm. Our framework also addresses the challenge of acquiring labeled training data by utilizing self-supervised learning. We evaluated the proposed framework on three public datasets and observed that it outperforms baseline models, indicating the feasibility and effectiveness of our approach.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"178 ","pages":"Article 113335"},"PeriodicalIF":7.2,"publicationDate":"2025-05-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144155109","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Path-planning algorithm for small environmental surveillance unmanned surface vehicles","authors":"Zhenyang Wang,&nbsp;Ping Yang,&nbsp;Diju Gao,&nbsp;Chunteng Bao","doi":"10.1016/j.asoc.2025.113342","DOIUrl":"10.1016/j.asoc.2025.113342","url":null,"abstract":"<div><div>Ports are essential hub facilities that provide support for economic development. However, the construction, development, and operation of ports increase the risk of environmental pollution in marine areas. Small environmental surveillance unmanned surface vehicles (ESUSVs) are being deployed to monitor port environments and prevent pollution. This study proposes a bidirectional elastic force contraction algorithm (BEFCA) and a Lévy flight weighted whale optimization (LFWWOA) and BEFCA hybrid algorithm (LFWWOA-BEFCA) to solve the path planning problem of ESUSVs. BEFCA solves the slow convergence and unsmooth path-characteristic problem of the elastic force contraction algorithm (EFCA) by employing a bidirectional search strategy and ship kinematics to smoothen the turning points in the path, respectively. LFWWOA uses a Lévy flight-based strategy in the global exploration phase of the whale optimization algorithm (WOA) to increase the solution diversity and improves the global and local search performance by modifying the coefficient calculation method and adding adaptive weighting coefficients. Thirteen benchmark functions were selected for the LFWWOA optimization performance experiments and compared with other intelligence algorithms. The results demonstrate that the proposed algorithm achieved the best global performance. Therefore, LFWWOA was used to optimize the BEFCA parameters, which resulted in higher-quality planned paths. Simulation experiments of real scenarios and complex environments showed that the path lengths and algorithm runtimes of BEFCA and LFWWOA-BEFCA outperformed those of the state prediction rapidly exploring random trees (spRRT) and spRRT-informed algorithms, respectively. The planned paths are consistent with the motion characteristics of ESUSVs, which can be used directly for tracking. The findings of this study indicate that shorter travel paths can be planned for ESUSVs in harbors for environmental monitoring, effectively solving the difficulty of tracking the paths of ESUSVs, and reducing energy consumption during the travel process.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"178 ","pages":"Article 113342"},"PeriodicalIF":7.2,"publicationDate":"2025-05-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144123387","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Diffusion-based road defect detection model integrating edge information and efficient multi-scale convolution","authors":"Xueqiu Wang ,&nbsp;Huanbing Gao ,&nbsp;Zemeng Jia","doi":"10.1016/j.asoc.2025.113332","DOIUrl":"10.1016/j.asoc.2025.113332","url":null,"abstract":"<div><div>Roads are vital infrastructure components, and the prompt detection and repair of defects are critical for their longevity and safety. This paper introduces the Edge Efficient Multi-Scale Focusing Diffusion Network (EEFNet), a precise method for road defect detection. The Edge Information Enhancement Module (EIEM) accentuates crack contours while minimizing background noise. An Efficient Multi-Scale Convolution (EMSConv) is proposed. The EMSConv captures features across multiple scales, thereby enhancing model efficiency through reduced computational demands and parameter count. The Focusing Diffusion Pyramid Network (FDPN) collects and distributes context-rich features across various scales using a diffusion mechanism, thereby improving detection capabilities. Additionally, the Task Dynamic Align Detection Head (TADDH) facilitates parameter sharing among detection heads, which enhances classification and localization accuracy. EEFNet has demonstrated a 92.7 % accuracy rate at 126 FPS (Frames Per Second) on a road defect dataset and has proven robustness on several diverse datasets including Unmanned Aerial Vehicle Asphalt Pavement Distress Dataset (UAPD), Visual Object Classes 2007 (VOC2007), Global Road Damage Detection2022 (GRDD2022), and Vision Meets Drone 2019 (Visdrone2019). In addition, by pruning the model and deploying it onto edge computing devices, practical experiments have demonstrated that the EEFNet model has substantial practical application value.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"178 ","pages":"Article 113332"},"PeriodicalIF":7.2,"publicationDate":"2025-05-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144124062","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Signature kernel ridge regression time series model: A novel approach for hydrological drought modeling using multi-station meteorological drought information","authors":"Mir Jafar Sadegh Safari ,&nbsp;Shervin Rahimzadeh Arashloo ,&nbsp;Babak Vaheddoost","doi":"10.1016/j.asoc.2025.113343","DOIUrl":"10.1016/j.asoc.2025.113343","url":null,"abstract":"<div><div>In the context of growing environmental challenges and the need for sustainable water resource management, hydrological drought prediction has gained prominence as a critical issue. Existing artificial intelligence and time series-based models for hydrological drought indices have traditionally been established using streamflow data. This study gives a significant progress in hydrological drought modeling through the introduction of the Signature Kernel Ridge Regression (SKRR) time series model. Instead of directly using rainfall and runoff data to develop a rainfall-runoff (RR) model, the Standardized Precipitation Evapotranspiration Index (SPEI) values in neighbor meteorological stations serve as inputs for estimating the Streamflow Drought Index (SDI) in target hydrometric stations, considering the 3-, 6-, and 12-month moving average time windows. The objective of this study is to enhance hydrological drought modeling by integrating soft computing techniques that effectively handle multivariate and irregular time series. The efficacy of the SKRR is compared with the well-established Generalized Regression Neural Network (GRNN), Random Forest (RF), and Auto Regressive Integrated Moving Average model with eXogenous input (ARIMAX). The findings indicate that SKRR is capable of precisely estimating SDI in three hydrometric stations using meteorological drought information from 14 stations, outperforming the GRNN, RF and ARIMAX models. The enhanced performance of the SKRR time series model stems from the utilization of a new and effective <em>signature kernel</em> which can be utilized for the study of irregularly sampled, multivariate time series in addition to be applicable to time series of different temporal spans while being a positive-definite kernel, facilitating usage in the Hilbert space. The novel drought based-RR model established by SKRR utilized various external stations’ meteorological drought indices to compute the hydrological drought indices in target stations not only enhances the modeling capability but also progress our understanding of drought dynamics by showcasing the power of soft computing in handling environmental uncertainty. Furthermore, it offers visions for developing of adaptive and resilience strategies to lessen the hazards caused by drought phenomenon.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"178 ","pages":"Article 113343"},"PeriodicalIF":7.2,"publicationDate":"2025-05-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144155114","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"TAS-TsC: A data-driven framework for Estimating Time of Arrival using Temporal-Attribute-Spatial Tri-space Coordination of truck trajectories","authors":"Mengran Li ,&nbsp;Junzhou Chen ,&nbsp;Guanying Jiang ,&nbsp;Fuliang Li ,&nbsp;Ronghui Zhang ,&nbsp;Siyuan Gong ,&nbsp;Zhihan Lv","doi":"10.1016/j.asoc.2025.113214","DOIUrl":"10.1016/j.asoc.2025.113214","url":null,"abstract":"<div><div>Accurately estimating the time of arrival (ETA) for trucks is crucial for optimizing transportation efficiency in logistics. GPS trajectory data provides valuable information for ETA, but challenges arise due to temporal sparsity, variable sequence lengths, and the interdependencies among multiple trucks. To address these issues, we propose the Temporal-Attribute-Spatial Tri-space Coordination (TAS-TsC) framework, which leverages three feature spaces – temporal, attribute, and spatial – to enhance ETA. Our framework consists of a Temporal Learning Module (TLM) that uses state space models to capture temporal dependencies, an Attribute Extraction Module (AEM) that transforms sequential features into structured attribute embeddings, and a Spatial Fusion Module (SFM) that models the interactions among multiple trajectories using graph representation learning. These modules collaboratively learn trajectory embeddings, which are then used by a Downstream Prediction Module (DPM) to estimate arrival times. We validate TAS-TsC on real truck trajectory datasets collected from Shenzhen, China, demonstrating its superior performance compared to existing methods.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"178 ","pages":"Article 113214"},"PeriodicalIF":7.2,"publicationDate":"2025-05-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144115338","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}