Engineering Applications of Artificial Intelligence最新文献_第10页

Efficient one-step multi-trial electroencephalograph spectral clustering via unsupervised covariance-based representations 通过基于协方差的无监督表示进行高效的一步式多试次脑电图频谱聚类

IF 7.5 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2024-10-21 DOI: 10.1016/j.engappai.2024.109502

As an important research branch of artificial intelligence, decoding motor imagery electroencephalograph (MI-EEG) is notoriously famous in engineering of constructing noninvasive brain-computer interfaces (BCIs). Clustering becomes a crucial manner in decoding MI-EEG due to lack of effective labels. However, recently clustering methods for EEG rely on modeling time-series characteristics with high dimensions, as well as classical clustering frameworks, which requires a large iterative consumption. To address these challenges, we proposed a novel Efficient one-step EEG spectral clustering (EosEEGsc) method for multi-trial scenarios. Firstly, two forms of covariance-base representations are constructed for the multi-trial MI-EEG samples using unsupervised manner. Subsequently, the similarity graphs are constructed according to such representation, and a weighting strategy between similarity graphs and spectral embedding is progressively iterated using a one-step spectral clustering manner. Comparative experiments were conducted on ten MI-EEG datasets from BCI Competitions. The EosEEGsc achieved better clustering performance with lower time complexity quickly converged to local optima during the one-step framework. Ablation studies have demonstrated the necessity of two key components of EosEEGsc, and parameter sensitivities have validated the robustness. Our method offers a novel option for online MI-BCIs. When labels for MI tasks cannot be quickly annotated, employing the EosEEGsc method enables rapid cluster acquisition, thereby guiding precise control instructions for MI-BCIs output.

作为人工智能的一个重要研究分支，运动图像脑电图（MI-EEG）解码在构建无创脑机接口（BCI）的工程中声名显赫。由于缺乏有效的标签，聚类成为解码 MI-EEG 的重要方式。然而，最近的脑电图聚类方法依赖于高维度的时间序列特征建模，以及经典的聚类框架，这需要大量的迭代消耗。针对这些挑战，我们提出了一种适用于多试验场景的新型高效一步式脑电图频谱聚类（EosEEGsc）方法。首先，采用无监督方式为多试验 MI-EEG 样本构建两种形式的协方差基础表示。随后，根据这种表示法构建相似性图，并采用一步光谱聚类的方式逐步迭代相似性图和光谱嵌入之间的加权策略。在 BCI 竞赛的十个 MI-EEG 数据集上进行了对比实验。在一步式框架中，EosEEGsc 以更低的时间复杂度快速收敛到局部最优，实现了更好的聚类性能。消融研究证明了 EosEEGsc 两个关键组件的必要性，而参数灵敏度则验证了其鲁棒性。我们的方法为在线 MI-BCI 提供了一种新的选择。当无法快速标注心肌缺血任务的标签时，采用 EosEEGsc 方法可以快速获取集群，从而为心肌缺血-脑干神经接口输出的精确控制指令提供指导。

{"title":"Efficient one-step multi-trial electroencephalograph spectral clustering via unsupervised covariance-based representations","authors":"","doi":"10.1016/j.engappai.2024.109502","DOIUrl":"10.1016/j.engappai.2024.109502","url":null,"abstract":"<div><div>As an important research branch of artificial intelligence, decoding motor imagery electroencephalograph (MI-EEG) is notoriously famous in engineering of constructing noninvasive brain-computer interfaces (BCIs). Clustering becomes a crucial manner in decoding MI-EEG due to lack of effective labels. However, recently clustering methods for EEG rely on modeling time-series characteristics with high dimensions, as well as classical clustering frameworks, which requires a large iterative consumption. To address these challenges, we proposed a novel <strong>E</strong>fficient <strong>o</strong>ne-<strong>s</strong>tep <strong>EEG s</strong>pectral <strong>c</strong>lustering (EosEEGsc) method for multi-trial scenarios. Firstly, two forms of covariance-base representations are constructed for the multi-trial MI-EEG samples using unsupervised manner. Subsequently, the similarity graphs are constructed according to such representation, and a weighting strategy between similarity graphs and spectral embedding is progressively iterated using a one-step spectral clustering manner. Comparative experiments were conducted on ten MI-EEG datasets from BCI Competitions. The EosEEGsc achieved better clustering performance with lower time complexity quickly converged to local optima during the one-step framework. Ablation studies have demonstrated the necessity of two key components of EosEEGsc, and parameter sensitivities have validated the robustness. Our method offers a novel option for online MI-BCIs. When labels for MI tasks cannot be quickly annotated, employing the EosEEGsc method enables rapid cluster acquisition, thereby guiding precise control instructions for MI-BCIs output.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":7.5,"publicationDate":"2024-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142534588","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Three-stage unsupervised learning approach fusing novel pseudo-label diffusion and math-physics translating for real-time structural damage detection 融合新型伪标签扩散和数学物理转换的三阶段无监督学习法，用于实时结构损伤检测

IF 7.5 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2024-10-21 DOI: 10.1016/j.engappai.2024.109438

Unsupervised learning techniques have been asserted as pivotal approaches in vibration-based structural damage detection due to scarce labeled data, whereas unrealistic premise on sufficient data from health status and inexplicit reflection on math-physics in various algorithms degrade their applicability in practical engineering. Therefore, this study devised a three-stage unsupervised learning approach for real-time structural damage detection. It incorporates novel pseudo-label diffusion based on different damage-sensitive-features extraction strategies, and math-physics translating using adaptative fuzzy clustering algorithm optimization. Comprehensive validations upon a well-known numerical benchmark model and full-scale laboratory shaking table tests are conducted, the results of which confirm the effectiveness and superiority of the proposed method. Integrated with novel pseudo-label diffusion and explicit math-physics translating mechanisms, the sophisticated framework is expected to fully automatically discriminate unequivocal structural damage in a strictly unsupervised schema, providing an explicit interpretation avenue from non-physical clustering to end-to-end decision-making.

由于标注数据稀缺，无监督学习技术被认为是基于振动的结构损伤检测的关键方法，而不切实际的健康状况充足数据前提和各种算法对数学物理的不明确反映降低了其在实际工程中的适用性。因此，本研究设计了一种用于实时结构损伤检测的三阶段无监督学习方法。该方法结合了基于不同损伤敏感特征提取策略的新型伪标签扩散，以及使用自适应模糊聚类算法优化的数学物理转换。对一个著名的数值基准模型和全尺寸实验室振动台试验进行了全面验证，结果证实了所提方法的有效性和优越性。与新颖的伪标签扩散和明确的数学物理转换机制相结合，这个复杂的框架有望在严格的无监督模式下全自动判别明确的结构损伤，提供从非物理聚类到端到端决策的明确解释途径。

{"title":"Three-stage unsupervised learning approach fusing novel pseudo-label diffusion and math-physics translating for real-time structural damage detection","authors":"","doi":"10.1016/j.engappai.2024.109438","DOIUrl":"10.1016/j.engappai.2024.109438","url":null,"abstract":"<div><div>Unsupervised learning techniques have been asserted as pivotal approaches in vibration-based structural damage detection due to scarce labeled data, whereas unrealistic premise on sufficient data from health status and inexplicit reflection on math-physics in various algorithms degrade their applicability in practical engineering. Therefore, this study devised a three-stage unsupervised learning approach for real-time structural damage detection. It incorporates novel pseudo-label diffusion based on different damage-sensitive-features extraction strategies, and math-physics translating using adaptative fuzzy clustering algorithm optimization. Comprehensive validations upon a well-known numerical benchmark model and full-scale laboratory shaking table tests are conducted, the results of which confirm the effectiveness and superiority of the proposed method. Integrated with novel pseudo-label diffusion and explicit math-physics translating mechanisms, the sophisticated framework is expected to fully automatically discriminate unequivocal structural damage in a strictly unsupervised schema, providing an explicit interpretation avenue from non-physical clustering to end-to-end decision-making.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":7.5,"publicationDate":"2024-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142534589","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Small object detection by Edge-aware Neural Network 利用边缘感知神经网络检测小物体

IF 7.5 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2024-10-21 DOI: 10.1016/j.engappai.2024.109406

The object detection method is widely applied in industrial inspections. However, many detectors face challenges in accurately capturing the blurred edge details of small objects, resulting in inaccurate bounding box predictions. To address this, we propose an Edge-aware Neural Network (EANN) for small object detection. Firstly, we introduce a Channel and Spatial Attention Fusion Module (CSAFM) to enhance the edge features of small objects, enabling the network to extract more discriminative information. Next, we propose a Multiple Aggregation Feature Pyramid (MAFP) to integrate multi-scale deep features into shallow features. This fusion enriches the shallow features with abundant semantic information, thereby aiding in the detection of small objects. Additionally, we propose a Side and Center Point Aligned Intersection over Union loss (SCPAIoULoss) to enhance the bounding box regression when there is minimal overlap between predicted and ground truth boxes. SCPAIoULoss combines Side Ratio (SR) loss, Center Point Distance (CPD) loss, and Intersection over Union (IoU) loss. The utilization of SR Loss directly constrains the width and height regression of bounding boxes, while CPD loss introduces stricter constraints to facilitate bounding box regression. Furthermore, IoU loss promotes the overall regression of predicted boxes. We extensively experiment on Tiny CityPersons, WiderFace, and our proposed dataset of base station data centers to validate the effectiveness of our method. The results indicate that our method surpasses several State-of-The-Art (SOTA) approaches in small object detection and can be effectively applied to the task of inspecting base station data centers.

物体检测方法被广泛应用于工业检测领域。然而，许多检测器在准确捕捉小物体模糊边缘细节方面面临挑战，导致边界框预测不准确。针对这一问题，我们提出了一种边缘感知神经网络（EANN）用于小物体检测。首先，我们引入了通道和空间注意力融合模块（CSAFM）来增强小物体的边缘特征，使网络能够提取更多的判别信息。接下来，我们提出了多重聚合特征金字塔（MAFP），将多尺度深度特征整合到浅层特征中。这种融合为浅层特征提供了丰富的语义信息，从而有助于小物体的检测。此外，我们还提出了边点和中心点对齐交集联合损失（SCPAIoULoss），用于在预测框和地面实况框重叠最小的情况下增强边界框回归。SCPAIoULoss 结合了边距损失 (SR)、中心点距离损失 (CPD) 和联合交叉损失 (IoU)。SR 损失的使用直接限制了边界框的宽度和高度回归，而 CPD 损失则引入了更严格的限制以促进边界框回归。此外，IoU 损失还能促进预测框的整体回归。我们在 Tiny CityPersons、WiderFace 和我们提出的基站数据中心数据集上进行了广泛实验，以验证我们方法的有效性。结果表明，我们的方法在小物体检测方面超越了几种最新方法（SOTA），可以有效地应用于基站数据中心的检测任务。

{"title":"Small object detection by Edge-aware Neural Network","authors":"","doi":"10.1016/j.engappai.2024.109406","DOIUrl":"10.1016/j.engappai.2024.109406","url":null,"abstract":"<div><div>The object detection method is widely applied in industrial inspections. However, many detectors face challenges in accurately capturing the blurred edge details of small objects, resulting in inaccurate bounding box predictions. To address this, we propose an Edge-aware Neural Network (EANN) for small object detection. Firstly, we introduce a Channel and Spatial Attention Fusion Module (CSAFM) to enhance the edge features of small objects, enabling the network to extract more discriminative information. Next, we propose a Multiple Aggregation Feature Pyramid (MAFP) to integrate multi-scale deep features into shallow features. This fusion enriches the shallow features with abundant semantic information, thereby aiding in the detection of small objects. Additionally, we propose a Side and Center Point Aligned Intersection over Union loss (SCPAIoULoss) to enhance the bounding box regression when there is minimal overlap between predicted and ground truth boxes. SCPAIoULoss combines Side Ratio (SR) loss, Center Point Distance (CPD) loss, and Intersection over Union (IoU) loss. The utilization of SR Loss directly constrains the width and height regression of bounding boxes, while CPD loss introduces stricter constraints to facilitate bounding box regression. Furthermore, IoU loss promotes the overall regression of predicted boxes. We extensively experiment on Tiny CityPersons, WiderFace, and our proposed dataset of base station data centers to validate the effectiveness of our method. The results indicate that our method surpasses several State-of-The-Art (SOTA) approaches in small object detection and can be effectively applied to the task of inspecting base station data centers.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":7.5,"publicationDate":"2024-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142534590","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Brain tumor classification using weighted least square twin support vector machine with fuzzy hyperplane 使用带模糊超平面的加权最小平方孪生支持向量机进行脑肿瘤分类

IF 7.5 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2024-10-21 DOI: 10.1016/j.engappai.2024.109450

Brain tumor is an aberrant growth of cells in the brain and represents one of the most lethal cancers around the world. The advanced machine learning models, like twin support vector machine have effectively addressed brain tumor classification tasks with promising results. However, despite its success, it lacks efficient learning as it solves a pair of quadratic programming problems and struggles to distinguish between support vectors and noises. To address these challenges, a novel multi-class classification model based on least square twin support vector machine and fuzzy concepts is formulated. It uses both membership and non-membership weights and integrates local neighborhood information among data points according to their importance. Moreover, to capture the uncertainty in the dataset, the proposed method computes a fuzzy hyperplane, taking all the parameters as fuzzy variables. Further, the model’s efficiency is enhanced by solving a system of linear equations only rather than solving a quadratic programming problem. To show the effectiveness of the proposed algorithm, the numerical experiments on 15 benchmark datasets in terms of average accuracy,

F

-measure, and training time are illustrated. The findings shows that the proposed technique outperforms other baseline models by achieving average accuracy of 88.79% with a linear kernel and 91.71% with a non-linear kernel. The proposed method is also applied to classify brain tumors into four different classes, achieving an average accuracy of 93.45%, which proves its outstanding performance. Moreover, the Friedman and Wilcoxon signed-rank statistical tests are used to confirm the method’s robustness and generalization capability.

脑肿瘤是脑部细胞的异常生长，是全球最致命的癌症之一。双支持向量机等先进的机器学习模型有效地解决了脑肿瘤分类任务，并取得了可喜的成果。然而，尽管它取得了成功，却缺乏高效的学习能力，因为它需要解决一对二次编程问题，并且难以区分支持向量和噪声。为了应对这些挑战，我们提出了一种基于最小平方孪生支持向量机和模糊概念的新型多类分类模型。该模型同时使用成员权重和非成员权重，并根据数据点的重要性整合数据点之间的局部邻域信息。此外，为了捕捉数据集中的不确定性，该方法将所有参数作为模糊变量，计算出一个模糊超平面。此外，通过只求解线性方程组而不是求解二次编程问题，提高了模型的效率。为了证明所提算法的有效性，我们对 15 个基准数据集进行了数值实验，从平均准确率、F-measure 和训练时间等方面进行了说明。实验结果表明，所提出的技术优于其他基准模型，使用线性内核时的平均准确率达到 88.79%，使用非线性内核时的平均准确率达到 91.71%。该方法还被用于将脑肿瘤分为四个不同的类别，平均准确率达到 93.45%，证明了其出色的性能。此外，Friedman 和 Wilcoxon 符号秩统计检验也证实了该方法的鲁棒性和泛化能力。

{"title":"Brain tumor classification using weighted least square twin support vector machine with fuzzy hyperplane","authors":"","doi":"10.1016/j.engappai.2024.109450","DOIUrl":"10.1016/j.engappai.2024.109450","url":null,"abstract":"<div><div>Brain tumor is an aberrant growth of cells in the brain and represents one of the most lethal cancers around the world. The advanced machine learning models, like twin support vector machine have effectively addressed brain tumor classification tasks with promising results. However, despite its success, it lacks efficient learning as it solves a pair of quadratic programming problems and struggles to distinguish between support vectors and noises. To address these challenges, a novel multi-class classification model based on least square twin support vector machine and fuzzy concepts is formulated. It uses both membership and non-membership weights and integrates local neighborhood information among data points according to their importance. Moreover, to capture the uncertainty in the dataset, the proposed method computes a fuzzy hyperplane, taking all the parameters as fuzzy variables. Further, the model’s efficiency is enhanced by solving a system of linear equations only rather than solving a quadratic programming problem. To show the effectiveness of the proposed algorithm, the numerical experiments on 15 benchmark datasets in terms of average accuracy, <span><math><mi>F</mi></math></span>-measure, and training time are illustrated. The findings shows that the proposed technique outperforms other baseline models by achieving average accuracy of 88.79% with a linear kernel and 91.71% with a non-linear kernel. The proposed method is also applied to classify brain tumors into four different classes, achieving an average accuracy of 93.45%, which proves its outstanding performance. Moreover, the Friedman and Wilcoxon signed-rank statistical tests are used to confirm the method’s robustness and generalization capability.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":7.5,"publicationDate":"2024-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142535245","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Robust autoregressive bidirectional gated recurrent units model for short-term power forecasting 用于短期电力预测的鲁棒自回归双向门控递归单元模型

IF 7.5 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2024-10-20 DOI: 10.1016/j.engappai.2024.109453

Accurate short-term power forecasting (STPF) provides reliable support for the stable operation of power systems. However, due to the randomness of consumer behavior and energy properties, outliers inevitably exist in power series. Considering its negative influence, effectively extracting features from the power series with outliers has become a significant challenge in STPF. This paper develops a robust hybrid model to handle this issue. The proposed model utilizes the robust regression technique to handle outliers. An adaptive rescaled Huber loss is developed to approximate the complex distribution of the actual power series. Moreover, the proposed model applies autoregressive and bidirectional gated recurrent units to extract linear and nonlinear features of power series, respectively. Meanwhile, the attention mechanism extracts the temporal feature through the attention representation, which considers the correlations between different moments. The proposed model obtains the optimal coefficients of determination between predictions and observations on the wind power series as 0.9629 and power load series as 0.978, which indicates that the proposed model performs competitive robustness and generalization on the daily operation of renewable energy systems.

准确的短期功率预测（STPF）为电力系统的稳定运行提供了可靠的支持。然而，由于用户行为和能源属性的随机性，电力序列中不可避免地存在异常值。考虑到其负面影响，如何有效地从带有异常值的电力序列中提取特征已成为 STPF 面临的重大挑战。本文开发了一种稳健混合模型来解决这一问题。所提出的模型利用稳健回归技术来处理异常值。开发了一种自适应重标定 Huber 损失，以近似实际幂级数的复杂分布。此外，该模型还应用了自回归和双向门控递归单元，分别提取功率序列的线性和非线性特征。同时，注意力机制通过注意力表征提取时间特征，该表征考虑了不同时刻之间的相关性。所提出的模型在风力发电序列和电力负荷序列上分别获得了 0.9629 和 0.978 的预测与观测之间的最优决定系数，这表明所提出的模型在可再生能源系统的日常运行中具有良好的鲁棒性和普适性。

{"title":"Robust autoregressive bidirectional gated recurrent units model for short-term power forecasting","authors":"","doi":"10.1016/j.engappai.2024.109453","DOIUrl":"10.1016/j.engappai.2024.109453","url":null,"abstract":"<div><div>Accurate short-term power forecasting (STPF) provides reliable support for the stable operation of power systems. However, due to the randomness of consumer behavior and energy properties, outliers inevitably exist in power series. Considering its negative influence, effectively extracting features from the power series with outliers has become a significant challenge in STPF. This paper develops a robust hybrid model to handle this issue. The proposed model utilizes the robust regression technique to handle outliers. An adaptive rescaled Huber loss is developed to approximate the complex distribution of the actual power series. Moreover, the proposed model applies autoregressive and bidirectional gated recurrent units to extract linear and nonlinear features of power series, respectively. Meanwhile, the attention mechanism extracts the temporal feature through the attention representation, which considers the correlations between different moments. The proposed model obtains the optimal coefficients of determination between predictions and observations on the wind power series as 0.9629 and power load series as 0.978, which indicates that the proposed model performs competitive robustness and generalization on the daily operation of renewable energy systems.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":7.5,"publicationDate":"2024-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142535244","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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}

引用次数: 0

Lane detection via disentangled representation network with slope consistency loss 通过具有斜率一致性损失的分解表示网络进行车道检测

IF 7.5 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2024-10-20 DOI: 10.1016/j.engappai.2024.109449

Existing works in lane detection focus on learning the general robust representation across different scenarios to overcome the impact of the lack of visual cues. However, factors leading to the absence of visual cues vary across different scenarios and the training data from challenging conditions is relatively small compared to common conditions. These problems result in the inability of existing methods to maintain robust lane detection in different scenarios for practical applications. To address these problems, this work presents a novel Disentangled Representation Network called DRNet, which disentangles the lane feature representations using a disentangled representation network to efficiently learn the lane representations corresponding to the specific condition. Meanwhile, DRNet also mitigates the adverse effects of data imbalance. Specifically, we disentangle lane representation via five branches, respectively to the common scenes, crowded objects, low light, dazzle light and other conditions. Due to the separated model of different conditions, each branch can be represented using a small number of parameters, which can be sufficiently learned using corresponding training subset. Moreover, existing works perform lane classification or regression using pixel-level losses, which neglect the important shape information. To this end, we design a novel slope consistency loss to take both global and local slope consistencies between prediction and ground truth into account for lane detection, which can adaptively adjust the lane shape and location. Extensive experiments on the CULane and TuSimple datasets show that our DRNet outperforms state-of-the-art methods, as it can reach 81.07% F1 on CULane and 97.97% on TuSimple.

车道检测方面的现有研究侧重于学习不同场景下的通用鲁棒表示法，以克服缺乏视觉线索的影响。然而，导致视觉线索缺失的因素在不同场景下各不相同，而且与普通情况相比，挑战性条件下的训练数据相对较少。这些问题导致现有方法无法在实际应用的不同场景中保持稳健的车道检测。为了解决这些问题，本研究提出了一种名为 DRNet 的新型分离表征网络，它利用分离表征网络对车道特征表征进行分离，从而有效地学习与特定条件相对应的车道表征。同时，DRNet 还能减轻数据不平衡带来的不利影响。具体来说，我们通过五个分支来分解车道表征，分别是常见场景、拥挤物体、弱光、眩光和其他条件。由于将不同条件的模型分离开来，每个分支都可以使用少量参数来表示，而这些参数可以通过相应的训练子集进行充分学习。此外，现有研究使用像素级损失进行车道分类或回归，忽略了重要的形状信息。为此，我们设计了一种新颖的斜率一致性损失，在车道检测中同时考虑预测值与地面实况之间的全局和局部斜率一致性，从而可以自适应地调整车道形状和位置。在 CULane 和 TuSimple 数据集上进行的大量实验表明，我们的 DRNet 优于最先进的方法，在 CULane 和 TuSimple 数据集上的 F1 分别达到 81.07% 和 97.97%。

{"title":"Lane detection via disentangled representation network with slope consistency loss","authors":"","doi":"10.1016/j.engappai.2024.109449","DOIUrl":"10.1016/j.engappai.2024.109449","url":null,"abstract":"<div><div>Existing works in lane detection focus on learning the general robust representation across different scenarios to overcome the impact of the lack of visual cues. However, factors leading to the absence of visual cues vary across different scenarios and the training data from challenging conditions is relatively small compared to common conditions. These problems result in the inability of existing methods to maintain robust lane detection in different scenarios for practical applications. To address these problems, this work presents a novel Disentangled Representation Network called DRNet, which disentangles the lane feature representations using a disentangled representation network to efficiently learn the lane representations corresponding to the specific condition. Meanwhile, DRNet also mitigates the adverse effects of data imbalance. Specifically, we disentangle lane representation via five branches, respectively to the common scenes, crowded objects, low light, dazzle light and other conditions. Due to the separated model of different conditions, each branch can be represented using a small number of parameters, which can be sufficiently learned using corresponding training subset. Moreover, existing works perform lane classification or regression using pixel-level losses, which neglect the important shape information. To this end, we design a novel slope consistency loss to take both global and local slope consistencies between prediction and ground truth into account for lane detection, which can adaptively adjust the lane shape and location. Extensive experiments on the CULane and TuSimple datasets show that our DRNet outperforms state-of-the-art methods, as it can reach 81.07% <em>F1</em> on CULane and 97.97% on TuSimple.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":7.5,"publicationDate":"2024-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142535243","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Development of physics-guided neural network framework for acid-base treatment prediction using carbon dioxide-based tubular reactor 利用基于二氧化碳的管式反应器开发用于酸碱处理预测的物理引导神经网络框架

IF 7.5 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2024-10-20 DOI: 10.1016/j.engappai.2024.109500

Accurate acid-base treatment prediction is necessary to achieve the required yield, given the inherent complexity, high nonlinearity, and restricted availability of data samples; to address this challenge, a data-driven approach was developed. However, the technique is constrained by the need for sufficient data to construct an accurate model and lacks both process insight and physical consistency. Therefore, this study introduces a physics-guided neural network model for acid-base treatment prediction in a dynamic tubular reactor using the fundamental physical intermediate variables obtained through the derivation process of the reaction schematic. By integrating batch experimental data, which provides key intermediate variables such as residence time and hydroxide ion concentration, the model addresses the challenge of high nonlinearity and limited data availability. The result shows that the physics-guided potential of a hydrogen predictor had outstanding performance in terms of prediction accuracy (greatest coefficient of determination value of 0.9381). The proposed model demonstrated an average improvement of 24.92% in pH prediction accuracy compared to traditional models without physical guidance, with a maximum improvement of up to 64.95% under limited data conditions. Moreover, downsampling tests revealed that the proposed model maintained robust performance with minimal accuracy reduction even when data was limited without overfitting implication.

鉴于酸碱处理本身的复杂性、高度非线性和数据样本的有限性，要达到所需的产量，就必须进行准确的酸碱处理预测；为应对这一挑战，我们开发了一种数据驱动方法。然而，由于需要足够的数据来构建准确的模型，该技术受到了限制，并且缺乏对工艺的深入了解和物理一致性。因此，本研究引入了一个物理引导的神经网络模型，利用通过反应原理图推导过程获得的基本物理中间变量，对动态管式反应器中的酸碱处理进行预测。通过整合批量实验数据（这些数据提供了关键的中间变量，如停留时间和氢氧根离子浓度），该模型解决了高非线性和数据可用性有限的难题。结果表明，氢气预测器的物理引导势能在预测准确性方面表现出色（最大决定系数值为 0.9381）。与没有物理指导的传统模型相比，所提出的模型在 pH 预测准确率方面平均提高了 24.92%，在数据有限的条件下，最高提高了 64.95%。此外，下采样测试表明，即使在数据有限的情况下，所提出的模型也能保持稳健的性能，准确率降低幅度极小，没有过拟合的影响。

{"title":"Development of physics-guided neural network framework for acid-base treatment prediction using carbon dioxide-based tubular reactor","authors":"","doi":"10.1016/j.engappai.2024.109500","DOIUrl":"10.1016/j.engappai.2024.109500","url":null,"abstract":"<div><div>Accurate acid-base treatment prediction is necessary to achieve the required yield, given the inherent complexity, high nonlinearity, and restricted availability of data samples; to address this challenge, a data-driven approach was developed. However, the technique is constrained by the need for sufficient data to construct an accurate model and lacks both process insight and physical consistency. Therefore, this study introduces a physics-guided neural network model for acid-base treatment prediction in a dynamic tubular reactor using the fundamental physical intermediate variables obtained through the derivation process of the reaction schematic. By integrating batch experimental data, which provides key intermediate variables such as residence time and hydroxide ion concentration, the model addresses the challenge of high nonlinearity and limited data availability. The result shows that the physics-guided potential of a hydrogen predictor had outstanding performance in terms of prediction accuracy (greatest coefficient of determination value of 0.9381). The proposed model demonstrated an average improvement of 24.92% in pH prediction accuracy compared to traditional models without physical guidance, with a maximum improvement of up to 64.95% under limited data conditions. Moreover, downsampling tests revealed that the proposed model maintained robust performance with minimal accuracy reduction even when data was limited without overfitting implication.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":7.5,"publicationDate":"2024-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142534591","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Perturbation defense ultra high-speed weak target recognition 扰动防御超高速弱目标识别

IF 7.5 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2024-10-20 DOI: 10.1016/j.engappai.2024.109420

Ultra high-speed target recognition in complex electromagnetic environments is a critical and fundamental machine perception issue. It is difficult to ensure privacy protection and intelligent confrontation with centralized training in many practical situations. In this paper, an efficient ultra high-speed target recognition approach, named InVision, for robot systems is proposed using deep trustworthy federated learning (DTFL). InVision is accurate, safe, fast, and robust. Particularly, geometric component transformer (GCT) is presented to significantly promote neural element's complex representation description ability. And an ambiguity-aware cooperative learning (AACL) scheme is developed to relieve the noisy label problem. Moreover, decentralized federated training (DFT) is designed to mitigate the intractable privacy protection problem, to efficiently search for similarities and reduce the representation redundancy. Furthermore, to promote the running speed of the system in real-world environments, a lightweight deep architecture, called Mobile-XB, is developed. Extensive quantitative and qualitative experiments are carried out, and the results demonstrate that InVision greatly outperforms the outstanding comparison methods, establishing efficient connections and extraction, and providing security guarantees.

复杂电磁环境下的超高速目标识别是机器感知的关键和基础问题。在很多实际情况下，集中训练很难确保隐私保护和智能对抗。本文利用深度可信联合学习（DTFL）为机器人系统提出了一种高效的超高速目标识别方法，命名为 InVision。InVision 具有准确、安全、快速和鲁棒性等特点。其中，几何分量变换器（GCT）的提出极大地提高了神经元的复杂表征描述能力。此外，还开发了一种模糊感知合作学习（AACL）方案来缓解噪声标签问题。此外，还设计了分散联合训练（DFT）来缓解难以解决的隐私保护问题，有效地搜索相似性并减少表示冗余。此外，为了提高系统在实际环境中的运行速度，还开发了一种名为 Mobile-XB 的轻量级深度架构。我们进行了广泛的定量和定性实验，结果表明 InVision 在建立高效连接和提取、提供安全保证等方面大大优于优秀的比较方法。

{"title":"Perturbation defense ultra high-speed weak target recognition","authors":"","doi":"10.1016/j.engappai.2024.109420","DOIUrl":"10.1016/j.engappai.2024.109420","url":null,"abstract":"<div><div>Ultra high-speed target recognition in complex electromagnetic environments is a critical and fundamental machine perception issue. It is difficult to ensure privacy protection and intelligent confrontation with centralized training in many practical situations. In this paper, an efficient ultra high-speed target recognition approach, named InVision, for robot systems is proposed using deep trustworthy federated learning (DTFL). InVision is accurate, safe, fast, and robust. Particularly, geometric component transformer (GCT) is presented to significantly promote neural element's complex representation description ability. And an ambiguity-aware cooperative learning (AACL) scheme is developed to relieve the noisy label problem. Moreover, decentralized federated training (DFT) is designed to mitigate the intractable privacy protection problem, to efficiently search for similarities and reduce the representation redundancy. Furthermore, to promote the running speed of the system in real-world environments, a lightweight deep architecture, called Mobile-XB, is developed. Extensive quantitative and qualitative experiments are carried out, and the results demonstrate that InVision greatly outperforms the outstanding comparison methods, establishing efficient connections and extraction, and providing security guarantees.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":7.5,"publicationDate":"2024-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142535242","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Reconstructing causal networks from data for the analysis, prediction, and optimization of complex industrial processes 从数据中重建因果网络，用于分析、预测和优化复杂的工业流程

IF 7.5 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2024-10-19 DOI: 10.1016/j.engappai.2024.109494

Lacking the understanding of the first principles leads to the apparent black box attributes of complex industrial processes. How to understand complex industrial processes from data and guiding industrial decision-making has become an urgent problem to solve. However, the existing data-driven models are also black boxes, focusing only on the correlation relationships between data without reflecting causal relationships. Therefore, this study addresses the challenge of double black boxes in complex industrial decision-making, proposing a research framework of "causal analysis → performance prediction → process optimization". Firstly, nonparametric copula entropy, network deconvolution, and information geometric causal inference are integrated to construct the causal relations network. Also, the observability and controllability of complex industrial processes are analyzed to provide valuable insights for improving the dataset. Then, drawing inspiration from the transformational machine learning idea, an explainable predictive model is constructed for predicting key performance indicators. Lastly, taking this predictive model as the process surrogate model, the optimal process parameters are solved using the particle swarm optimization algorithm. Moreover, the dataset of 16600 samples from a real-world injection molding process is used for application validation. The research results show that by reconstructing the causal relations network from data, the proposed framework can support the analysis, prediction, and optimization of complex industrial processes, achieving the decision-making goals of safety, robustness, improving quality and efficiency.

由于缺乏对第一性原理的理解，导致复杂工业过程具有明显的黑箱属性。如何从数据中理解复杂的工业过程，并指导工业决策，已成为亟待解决的问题。然而，现有的数据驱动模型也是黑箱，只关注数据之间的相关关系，而不反映因果关系。因此，本研究针对复杂工业决策中的双黑箱难题，提出了 "因果分析→性能预测→流程优化 "的研究框架。首先，综合运用非参数共轭熵、网络解卷积和信息几何因果推理等方法构建因果关系网络。同时，分析复杂工业流程的可观测性和可控性，为改进数据集提供有价值的见解。然后，从变革机器学习思想中汲取灵感，构建一个可解释的预测模型，用于预测关键性能指标。最后，将该预测模型作为过程代理模型，使用粒子群优化算法求解最佳过程参数。此外，还利用实际注塑过程中的 16600 个样本数据集进行了应用验证。研究结果表明，通过从数据中重建因果关系网络，所提出的框架可以支持复杂工业过程的分析、预测和优化，实现安全、稳健、提高质量和效率的决策目标。

{"title":"Reconstructing causal networks from data for the analysis, prediction, and optimization of complex industrial processes","authors":"","doi":"10.1016/j.engappai.2024.109494","DOIUrl":"10.1016/j.engappai.2024.109494","url":null,"abstract":"<div><div>Lacking the understanding of the first principles leads to the apparent black box attributes of complex industrial processes. How to understand complex industrial processes from data and guiding industrial decision-making has become an urgent problem to solve. However, the existing data-driven models are also black boxes, focusing only on the correlation relationships between data without reflecting causal relationships. Therefore, this study addresses the challenge of double black boxes in complex industrial decision-making, proposing a research framework of \"causal analysis → performance prediction → process optimization\". Firstly, nonparametric copula entropy, network deconvolution, and information geometric causal inference are integrated to construct the causal relations network. Also, the observability and controllability of complex industrial processes are analyzed to provide valuable insights for improving the dataset. Then, drawing inspiration from the transformational machine learning idea, an explainable predictive model is constructed for predicting key performance indicators. Lastly, taking this predictive model as the process surrogate model, the optimal process parameters are solved using the particle swarm optimization algorithm. Moreover, the dataset of 16600 samples from a real-world injection molding process is used for application validation. The research results show that by reconstructing the causal relations network from data, the proposed framework can support the analysis, prediction, and optimization of complex industrial processes, achieving the decision-making goals of safety, robustness, improving quality and efficiency.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":7.5,"publicationDate":"2024-10-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142534332","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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}

引用次数: 0

CNCAN: Contrast and normal channel attention network for super-resolution image reconstruction of crops and weeds CNCAN：用于农作物和杂草超分辨率图像重建的对比度和正常通道注意力网络

IF 7.5 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2024-10-19 DOI: 10.1016/j.engappai.2024.109487

Numerous studies have been performed to apply camera vision technologies in robot-based agriculture and smart farms. In particular, to obtain high accuracy, it is essential to procure high-resolution (HR) images, which requires a high-performance camera. However, due to high costs it is difficult to widely apply the camera in agricultural robots. To overcome this limitation, we propose contrast and normal channel attention network (CNCAN) for super-resolution reconstruction (SR), which is the first research for the accurate semantic segmentation of crops and weeds even with low-resolution (LR) images captured by low-cost and LR camera. Attention block and activation function that considers high frequency and contrast information of images are used in CNCAN, and the residual connection method is applied to improve the learning stability.

As a result of experimenting with three open datasets, namely, Bonirob, rice seedling and weed, and crop/weed field image (CWFID) datasets, the mean intersection of union (MIOU) results of semantic segmentation for crops and weeds with SR images through CNCAN were 0.7685, 0.6346, and 0.6931 in the Bonirob, rice seedling and weed, and CWFID datasets, respectively, confirming higher accuracy than other state-of-the-art methods for SR.

在基于机器人的农业和智能农场中应用相机视觉技术的研究不胜枚举。特别是，要获得高精度，必须获得高分辨率（HR）图像，这就需要高性能相机。然而，由于成本高昂，相机很难广泛应用于农业机器人。为了克服这一限制，我们提出了用于超分辨率重建（SR）的对比度和正常通道注意网络（CNCAN），这是首次针对低成本和低分辨率相机拍摄的低分辨率（LR）图像对农作物和杂草进行精确语义分割的研究。CNCAN 采用了考虑图像高频和对比度信息的注意块和激活函数，并应用残差连接法提高了学习稳定性。通过对 Bonirob、水稻秧苗和杂草以及作物/杂草田图像（CWFID）三个开放数据集的实验，CNCAN 利用 SR 图像对作物和杂草进行语义分割的平均交集联合（MIOU）结果在 Bonirob、水稻秧苗和杂草以及 CWFID 数据集中分别为 0.7685、0.6346 和 0.6931，证实了比其他最先进的 SR 方法更高的精度。

{"title":"CNCAN: Contrast and normal channel attention network for super-resolution image reconstruction of crops and weeds","authors":"","doi":"10.1016/j.engappai.2024.109487","DOIUrl":"10.1016/j.engappai.2024.109487","url":null,"abstract":"<div><div>Numerous studies have been performed to apply camera vision technologies in robot-based agriculture and smart farms. In particular, to obtain high accuracy, it is essential to procure high-resolution (HR) images, which requires a high-performance camera. However, due to high costs it is difficult to widely apply the camera in agricultural robots. To overcome this limitation, we propose contrast and normal channel attention network (CNCAN) for super-resolution reconstruction (SR), which is the first research for the accurate semantic segmentation of crops and weeds even with low-resolution (LR) images captured by low-cost and LR camera. Attention block and activation function that considers high frequency and contrast information of images are used in CNCAN, and the residual connection method is applied to improve the learning stability.</div><div>As a result of experimenting with three open datasets, namely, Bonirob, rice seedling and weed, and crop/weed field image (CWFID) datasets, the mean intersection of union (MIOU) results of semantic segmentation for crops and weeds with SR images through CNCAN were 0.7685, 0.6346, and 0.6931 in the Bonirob, rice seedling and weed, and CWFID datasets, respectively, confirming higher accuracy than other state-of-the-art methods for SR.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":null,"pages":null},"PeriodicalIF":7.5,"publicationDate":"2024-10-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142535247","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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}

引用次数: 0