Applied Soft Computing最新文献

{"title":"Target temperature field prediction via a thermodynamic knowledge-based artificial neural network","authors":"Jincheng Chen ,&nbsp;Feiding Zhu ,&nbsp;Yuge Han ,&nbsp;Dengfeng Ren","doi":"10.1016/j.asoc.2025.112972","DOIUrl":"10.1016/j.asoc.2025.112972","url":null,"abstract":"<div><div>With the rapid development of artificial intelligence, representation-supervised neural networks have been widely used in the fast solution of physical field. However, a large number of temperature prediction networks do not take environmental parameters into account, or only use parameters as simple input conditions, which greatly reduces the accuracy of their results. This paper proposes an accurate and low-cost method for adding the conditional parameters to intelligent prediction networks. A novel parameter encoder block is designed based on the heat transfer theory achieving thermodynamic knowledge-based parameter feature extraction. Meanwhile, an improved method for inputting time condition is proposed to characterize the temporal characteristics, which can reduce the requirement of dataset for transient temperature prediction, compared with LSTM. In addition, a thermal loss for temperature images is introduced to accelerate the convergence process in the model. Moreover, a CycleGAN-based temperature prediction network (CBTPN) is constructed for fast temperature prediction of a cube or different tanks. Temperature or infrared images predicted by our network exhibit MAE of less than 2.33 % and SSIM of more than 80.21 %. By embedding physical mechanisms into neural networks, this study this study pioneers a structured approach to refining physical parameters into thermodynamic knowledge-based signals for improved image generation, addressing the accuracy and efficiency limitations of data-driven algorithms caused by their insufficient understanding of parameter mechanisms. Finally, parameter cognitive evaluation proves that our approach can not only recognize the accurate semantics of heat transfer parameters, but also sense the meteorological laws.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"174 ","pages":"Article 112972"},"PeriodicalIF":7.2,"publicationDate":"2025-03-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143643858","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":"Saccade inspired Attentive Visual Patch Transformer for image sentiment analysis","authors":"Jing Zhang,&nbsp;Jixiang Zhu,&nbsp;Han Sun,&nbsp;Xinzhou Zhang,&nbsp;Jiangpei Liu","doi":"10.1016/j.asoc.2025.112963","DOIUrl":"10.1016/j.asoc.2025.112963","url":null,"abstract":"<div><div>The generation of image-evoked emotion is usually regarded as a transient process in the image sentiment analysis. However, according to the saccade mechanism of the human visual system, the evoked emotion generated during the saccade process changes over time and attention. Based on above analysis, we propose an Attentive Visual Patch Transformer (AVPT), using visual attention sequence to represent the sentiment context of images and predict the possible distribution of sentiment. In AVPT, the spatial structure in the form of patches are reconstructed and reorganized by visual attention shift sequentially. Simultaneously, the temporal characteristics of attention shift are introduced to the relative position encoding, and merged in a self-attention manner to form a spatial–temporal process similarly to the human visual system. Specifically, we propose a sequence attention shift module to simulate the saccade process, which obtains sequence attention and reduces the computational effort by group attentive convolutional gate recurrent unit. Then, a spatial–temporal correlation encoder module is proposed to encode temporal attention with spatial visual features and obtain the sequential visual features of saccade. Finally, a self-attention fusion module is used to extract the correlation hidden in the relative encoding features. Our proposed AVPT achieves excellent performance on visual sentiment distribution prediction and is comparable to state-of-the-art methods, as demonstrated by extensive experiments on the Flickr_LDL and Twitter_LDL datasets.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"174 ","pages":"Article 112963"},"PeriodicalIF":7.2,"publicationDate":"2025-03-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143629921","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":"Unmanned aerial vehicle takeoff point search algorithm with information sharing strategy of random trees for multi-area coverage task","authors":"Shouwen Yao,&nbsp;Xiaoyu Wang,&nbsp;Siqi Huang,&nbsp;Renjie Xu,&nbsp;Yinghua Zhao","doi":"10.1016/j.asoc.2025.112970","DOIUrl":"10.1016/j.asoc.2025.112970","url":null,"abstract":"<div><div>This study proposes a novel approach to optimize full-coverage search in distributed task areas using a single Unmanned Ground Vehicle (UGV) to deliver an Unmanned Aerial Vehicle (UAV) to the takeoff points of each task area along the shortest possible path. Unlike the traditional Traveling Salesman Problem (TSP), task areas are not fixed nodes, and obstacles must be considered. To address these challenges, a probability-based Rapid-exploration Random Tree (<em>p</em>-RRT) with an information-sharing strategy is introduced, significantly improving the efficiency of locating takeoff points in complex environments. A dual optimization method further reduces the number of nodes and path length planned by the D* algorithm, achieving up to an 80 % reduction in nodes and improving path efficiency. Additionally, a simulated annealing (SA) algorithm optimizes the connection sequence of takeoff points, reducing total path length by 35.05 % compared to the initial path and 22.66 % compared to the traditional Random Sampling Method (RSM). Experiments confirm that the proposed algorithms can effectively enhance UGV-UAV collaboration with reducing path complexity and improving energy efficiency, and thus streamline multi-area coverage tasks.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"174 ","pages":"Article 112970"},"PeriodicalIF":7.2,"publicationDate":"2025-03-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143610990","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":"Deep learning in produce perception of harvesting robots: A comprehensive review","authors":"Yuhao Jin ,&nbsp;Xiaoyu Xia ,&nbsp;Qizhong Gao ,&nbsp;Yong Yue ,&nbsp;Eng Gee Lim ,&nbsp;Prudence Wong ,&nbsp;Weiping Ding ,&nbsp;Xiaohui Zhu","doi":"10.1016/j.asoc.2025.112971","DOIUrl":"10.1016/j.asoc.2025.112971","url":null,"abstract":"<div><div>In recent years, the global demand for produce has surged, alongside labor shortages, driving the development of agricultural automation, particularly in harvesting robots. Deep learning-based computer vision algorithms have become key to produce perception, demonstrating significant potential. We systematically review the current application of deep learning in produce perception for harvesting robots, providing an in-depth analysis of existing public datasets, with a focus on 2D produce recognition and 3D produce localization. Furthermore, we review and analyze the existing algorithms, highlighting their limitations and challenges. In addition, we explore future research directions of deep learning in produce perception, aiming to promote the continued advancement and innovation of technologies in this area.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"174 ","pages":"Article 112971"},"PeriodicalIF":7.2,"publicationDate":"2025-03-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143620229","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Transferable adversarial attacks against face recognition using surrogate model fine-tuning","authors":"Yasmeen M. Khedr ,&nbsp;Xin Liu ,&nbsp;Haobo Lu ,&nbsp;Kun He","doi":"10.1016/j.asoc.2025.112983","DOIUrl":"10.1016/j.asoc.2025.112983","url":null,"abstract":"<div><div>Deep Neural Networks have significantly advanced Face Recognition performance yet remain susceptible to adversarial attacks, posing significant security and user privacy threats in real-world applications. In recent years, black box attacks have attracted wide attention to craft highly transferable adversarial examples by training surrogate models. However, most of these methods primarily depend on stealing knowledge by accessing the soft label from the target model using either synthetic training data or data free without awareness of the knowledge type, which can affect the improvement of transferability between the surrogate and the target models. Additionally, these attacks still need to improve the surrogate model’s accuracy without using many queries. To this end, we propose Tune2Transfer, a novel attack method that enhances adversarial transferability by fine-tuning the surrogate model with different types of knowledge with limited queries on the target model by the hard label only. Specifically, it collects a small face image dataset, considering the adversary’s limited knowledge. To overcome the challenge of knowledge type, Tune2Transfer imposes three sampling assumptions: clean images only, the perturbed images, or combining both, generating images on the surrogate model, and then feeding them to the target model to obtain the hard label. The perturbed images are generated by perturbing them using the Covariance Matrix Adaptation Evolution Strategy or Momentum Iteration Fast Gradient Sign Method. Besides, we leverage pre-trained models to fine-tune surrogate models to avoid large queries. In this way, we could leverage knowledge transferred from the target model, resulting in superior transferability. Extensive experiments conducted on two typical datasets demonstrate the efficacy of Tune2Transfer, increasing the attack success rates significantly.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"174 ","pages":"Article 112983"},"PeriodicalIF":7.2,"publicationDate":"2025-03-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143610995","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":"A semi-supervised non-negative matrix factorization model for scRNA-seq data analysis","authors":"Junjie Lan ,&nbsp;Xiaoling Zhuo ,&nbsp;Siman Ye ,&nbsp;Jin Deng","doi":"10.1016/j.asoc.2025.112982","DOIUrl":"10.1016/j.asoc.2025.112982","url":null,"abstract":"<div><div>Single-cell RNA sequencing (scRNA-seq) technology enables the measurement of cellular gene expression at the single-cell level, thus facilitating cell clustering at the gene level. Despite numerous dimensionality reduction methods developed for scRNA-seq data, many are limited to analyzing individual gene expression matrices and struggle to address false positives and false zero expression entries effectively. Moreover, existing methods often underutilize prior knowledge of similarity and dissimilarity between multi-omics data, leading to the loss of intercellular correlations and shared structural information, thus hindering desired dimensionality reduction outcomes. To address these limitations, a novel model termed joint non-negative matrix factorization with similarity and dissimilarity constraints (SDJNMF) was proposed to tailor for scRNA-seq data clustering. The model leverages prior knowledge of similarity and dissimilarity across multiple gene expression matrices, facilitating joint non-negative matrix factorization to extract common features from multi-omics data. By preserving shared structural and cellular relevance information, SDJNMF enhances the clustering of similar cells while effectively separating dissimilar ones. Furthermore, the SDJNMF model incorporates sparse Singular Value Decomposition during initialization to mitigate noise and redundancy and ensure robust dimensionality reduction. The experimental results demonstrate that the SDJNMF model exhibits superior performance on the 10 datasets, not only outperforming the other 14 algorithms in terms of clustering accuracy on the 9 datasets, but also enhancing the <span><math><mrow><mi>A</mi><mi>R</mi><mi>I</mi></mrow></math></span> of SDJNMF by an average of 0.0687 in comparison to the second-best algorithm on each dataset. In the visual representation, the model is able to efficiently and accurately cluster similar cells and effectively discriminate different classes of cells from each other. Additionally, the SDJNMF model was applied to identify informative genes and conduct enrichment analysis, validating that genes identified by SDJNMF significantly influence biological processes. Overall, the SDJNMF offers innovative tools for cell cluster identification and advances biological research. The source code of SDJNMF is available online at <span><span>https://github.com/Jindsmu/SDJNMF</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"174 ","pages":"Article 112982"},"PeriodicalIF":7.2,"publicationDate":"2025-03-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143629923","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":"Combination weighting method using Z-numbers for multi-criteria decision-making","authors":"Huan-Jyh Shyur","doi":"10.1016/j.asoc.2025.112992","DOIUrl":"10.1016/j.asoc.2025.112992","url":null,"abstract":"<div><div>This study introduces a hybrid approach to determining criteria weights in Multi-Criteria Decision Making (MCDM), combining subjective weight methods with objective weighting techniques based on Z-number theory. The methodology is applied in a practical context involving the establishment of a bank's call center data analysis platform. Leveraging the inherent uncertainty and reliability considerations in decision-making processes, the hybrid method offers a robust framework for decision support. Through empirical validation and case study analysis, the effectiveness of the proposed approach is demonstrated, highlighting its ability to balance theoretical robustness with practical applicability. The study underscores the importance of ongoing research in MCDM, particularly in developing innovative methods to address the complexities of decision-making environments. Insights from this research provide valuable guidance for practitioners and researchers seeking to enhance MCDM processes across diverse domains.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"174 ","pages":"Article 112992"},"PeriodicalIF":7.2,"publicationDate":"2025-03-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143620228","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":"Exploring coordinated motion patterns of facial landmarks for deepfake video detection","authors":"Yue Zhang ,&nbsp;Run Niu ,&nbsp;Xianlin Zhang ,&nbsp;Siqi Chen ,&nbsp;Mingdao Wang ,&nbsp;Xueming Li","doi":"10.1016/j.asoc.2025.112974","DOIUrl":"10.1016/j.asoc.2025.112974","url":null,"abstract":"<div><div>Due to the rich geometric and motion information they contain, recent studies indicate that facial landmark clues have significant potential for deepfake video detection. In this paper, we make a key observation that there exist coordinated motions among different facial landmarks for real individuals. While the forgery methods focus more on appearance realism, thus likely to disrupt the underlying coordinated motion patterns. Inspired by this observation, this paper explores how to leverage coordinated motion patterns among facial landmarks to enhance deepfake detection. First, we introduce a coordinated motion landmarks mining strategy (CMLMS), to effectively identify correlated landmarks. Utilizing these correlations, we propose a landmark temporal dynamic relation module (LTDRM), which focuses on the coordinated motion patterns between landmarks while extracting their spatiotemporal features. Specifically, LTDRM constructs an adjacency matrix based on the correlated landmarks and uses graph convolution to selectively aggregate information between correlated landmarks. Additionally, LTDRM is a plug-and-play module and can boost the performance of existing deepfake detection methods with minimal computational overhead. Experimental results validate the effectiveness and generalizability of our method.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"174 ","pages":"Article 112974"},"PeriodicalIF":7.2,"publicationDate":"2025-03-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143637198","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":"A nondominated sorting simplified swarm optimization with local search mechanisms for multi-objective vehicle routing problems with time windows","authors":"Chyh-Ming Lai ,&nbsp;Chun-Chih Chiu ,&nbsp;Tzu-Li Chen","doi":"10.1016/j.asoc.2025.112989","DOIUrl":"10.1016/j.asoc.2025.112989","url":null,"abstract":"<div><div>In addressing the complexities of modern logistics, this study introduces a novel multi-objective formulation for vehicle routing problems with time windows (MO-VRPTW), targeting minimizing travel distance, enhancing customer satisfaction, and equalizing driver workloads. We introduce an innovative hybrid multi-objective evolutionary algorithm (MOEA) leveraging nondominated sorting simplified swarm optimization to effectively merge the advantages of various optimization strategies. A key aspect of this advancement is the incorporation of the Lin−Kernighan<strong>−</strong>Helsgaun (LKH) heuristic, which delivers a superior initial solution, thereby markedly enhancing the speed of convergence. Additionally, we pioneered a local search method inspired by the A* algorithm designed to refine the search process's exploration and exploitation stages. Solomon's benchmark instances, a recognized standard in the VRPTW field, were used to validate our algorithm's effectiveness. Our algorithm demonstrated superior performance in addressing MO-VRPTW through meticulous statistical analysis, outperforming state-of-the-art algorithms, such as MOPSO, NSGA-II, MOEA/D, and SPEA2, regarding efficiency and solution diversity. This study not only advances algorithmic performance but also thoughtfully considers the interests of key supply chain stakeholders.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"174 ","pages":"Article 112989"},"PeriodicalIF":7.2,"publicationDate":"2025-03-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143642505","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":"Dual degradation image inpainting method via adaptive feature fusion and U-net network","authors":"Yuantao Chen ,&nbsp;Runlong Xia ,&nbsp;Kai Yang ,&nbsp;Ke Zou","doi":"10.1016/j.asoc.2025.113010","DOIUrl":"10.1016/j.asoc.2025.113010","url":null,"abstract":"<div><div>Most existing image inpainting methods are designed to address a single specific task, such as super-resolution, denoising, or colorization, with few models capable of handling dual degradation simultaneously. Moreover, current algorithms that tackle multiple image degradation problems often suffer from complex structures, prolonged training times, and high labor costs. In this paper, we propose a Dual Degradation Network via Adaptive Feature Fusion and U-Net (AFFU). The network employs a Self-Guided Module (SGM) to fuse multi-scale image information, effectively eliminating certain defects in the image. A coder-decoder module with null convolution is utilized to consolidate the semantic information of the image, enabling intermediate image colorization. Additionally, an Adaptive Multi-feature Fusion Module (AMF) and Information Transfer Mechanism (ITM) are introduced to link these two major structures, adaptively selecting and retaining image features during network progression to prevent the loss of useful information. Experimental results demonstrate that the proposed dual image degradation restoration network model, based on adaptive multi-feature fusion, achieves optimal visual generation. Evaluations on CelebA dataset and Landscape dataset show that the proposed method outperforms comparable approaches in terms of Structural Similarity (SSIM) and Learned Perceptual Image Patch Similarity (LPIPS).</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"174 ","pages":"Article 113010"},"PeriodicalIF":7.2,"publicationDate":"2025-03-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143620233","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}