Applied Soft Computing最新文献

Cervical cancer remains a significant global health concern. Given the disparity between limited medical resources and the requisite professional personnel, the coverage of cervical screening is inadequate, particularly in underdeveloped areas. Computer-assisted liquid-based cytology diagnostic systems offer favorable solutions.

Detection of small nuclei within a complex liquid-based environment poses a challenge, exacerbated by the restricted availability of manual annotations. In this study, we propose FuseDLAM, a comprehensive computer-aided diagnostic system, which employs enhanced YOLOv8 with transformers for rapid localization of individual squamous epithelial cells. We leverage artificial intelligence-generated content techniques for data augmentation, effectively reducing the need for costly manual annotations. By integrating multiple deep convolutional neural network models with self-attention mechanisms, the system extracts crucial features from cell nuclei. These features are then fused through a fully connected layer to facilitate robust cell classification. FuseDLAM achieves an F1-score of 99.3$\%$ on the public SIPaKMeD dataset, demonstrating comparability with state-of-the-art approaches. It also proves its practical applicability in real-world clinical scenarios, achieving an F1-score of 91.2 % in identifying abnormal cervical squamous cells. Additionally, ablation experiments in both datasets validate the model's effectiveness. This underscores its potential for widespread application in medical imaging tasks.

The Baltic Dry Index (BDI) is one of the leading indexes that is the most commonly used to reflect the prosperity of the shipping industry. The index’s volatility indicates the operational risks that shipping-related enterprises and service institutions may face. In order to more accurately estimate the volatility, this study proposes a secondary decomposition-ensemble model that can be used to predict interval-valued time series (ITS) of the BDI. Four main steps are involved, namely ITS construction and primary decomposition, secondary decomposition, component ITS forecasting, and ensemble. To be specific, bivariate empirical mode decomposition (BEMD) is employed for the primary decomposition, and multivariate variational mode decomposition (MVMD) is used for the secondary decomposition. Using daily BDI data, an empirical analysis is conducted to verify the proposed model. The investigation shows that, compared to other models, the proposed method has better forecasting performance and stronger robustness in ITS forecasting of the BDI. The results indicate that using the proposed model is a promising method for the volatility estimation of complex ITS data.

Monitoring heat events in dairy cows is crucial for determining the heat on time, and the heat events have usually been estimated using machine learning on cow behavioral data collected from wireless activity sensors recently. However, ensuring robust performance of heat detection is difficult because of the difference in data domains (e.g., sensor types) and insufficient heat-labeled data. Therefore, this study proposes a multi-autoencoder-based heat detection in dairy cows that can represent the common representation of cow behavior across the different sensors. The proposed method can train a sensor-type agnostic heat detector using entire labeled data from the two different sensor types by aligning the latent spaces for two sensors. In addition, our approach can train the model by combining anomaly detection and weakly supervised classification to improve the performance of heat detection that can reduce the dependency on label accuracy. The results showed that the proposed approach improved cow heat detection performance by approximately 46 % than independently trained autoencoders, and the average F1-score increased by up to 0.70. The proposed method also outperformed other supervised and unsupervised learning models in heat detection using our dataset. From the results, our model effectively estimates cow behaviors by integrating sensor modalities, thereby enhancing data capabilities in low-resource settings. This study can be key for addressing the detection discrepancy in time series data based on the location of the mounted sensor, and offers the advantage of practical applications to various activity sensors currently used on farms.

A reduced biquaternion neural network (RQNN) has achieved significant success in machine learning. However, as the reduced biquaternion algebra system contains infinite zero divisors, the RQNN can be easily trapped in a local minimum and overfitting. In this paper, we propose a new regularization scheme for the RQNN to address these issues. Firstly, we propose a new operation in the reduced biquaternion domain named the reduced biquaternion complex modulus (RQCM), which can extract the scale transformation of reduced biquaternions and decrease the unreasonable network constraints caused by constrained phases. Secondly, we mathematically analyse the properties of the reduced biquaternions and obtain the geometric meaning of the RQCM. Finally, we propose an improved weight decay method using the RQCM which can better project the reduced biquaternion in terms of the scale and phase. In addition, our proposed method can effectively solve the non-differentiability of reduced biquaternion matrix and overfitting problem in the process of network parameter updating. The experimental results demonstrate that the proposed method is effective in color image classification and denoising taskas, and outperforms the state of the arts.

To incorporate Local Energy Management System (LEMS) based Tertiary Controller (TC) operation with multiple Photovoltaic based Distributed Generations (PV-DGs) level Primary Controller/ PC: Independent DG Controllers (IDGC) more adequately, a new representation learning (Symmetrical Input-Output weight Encoded Data Compression/ SIODC, with Exponential Expansion based Random Vector Functional Link Network towards Softmax layer’s Control Reference Estimation/ ExRVFLN-SoCRE) based MPPT controller is proposed in this paper in terms of Secondary Controllers (SCs). A new Hybrid LEMS (HyLEMS) is presented towards the proposed Deep Neural Network based SCs (i.e. DNN-SCs) in a distributed (SC_dist), as well as centralized (SC_cent) manner, to ease the computational, and communication requirements. To investigate that multiple PV-DGs based duty cycle (kth instant Control References/CRs estimation) controllers are considered here with auxiliary Battery Energy Storage Systems (BESS), and AC utility, integrated to Common DC feeder in terms of DC-DC converter dynamics. To avoid Control Reference Estimation Error (CREE) due to initial randomization, optimization subroutines are incorporated for SIODC by generalized semi-supervised learning, and for ExRVFLN-SoCRE by Lagrange multiplier weighted, rms based cost function. The proposed control performance is verified in MATLAB Simulink® based average modeling, and validated through dSPACE DS1104 based RTI with multi-PV (emulators) test-bench as well.

Addressing the decision-making challenge arising from the uncertainty of human cognition, three-way decision (3WD) and q-rung orthopair fuzzy sets (q-ROFSs) are integrated in this paper to propose a multi-strategy three-way decision approach (MS3WDA) for tri-state risk loss (TSRL) under q-rung orthopair fuzzy environment. Based on the ternary thinking of human cognition, the risk loss with hesitation state is considered and constructed under q-rung orthopair fuzzy environment. The TSRL with hesitation state is further constructed by combining the q-rung orthopair fuzzy (q-ROF) information. The conditional probability adopted by the original object classes is improved and extended by the three components of q-ROFSs. Next, the TSRL with q-ROF information and three components of q-ROFSs are integrated with decision-theoretic rough sets (DTRSs) to establish a novel 3WD model. Some relevant properties are also analyzed and discussed for the developed 3WD model. Then, its multi-strategy decision method is proposed based on the multi-strategy perspective. The related strategies with five different levels are designed by considering three different risk appetite perspectives and four different aspects of q-ROF information. The relevant threshold theorems are also given and proved to further provide the theoretical support for our MS3WDA. According to the five different strategies, we further derive the corresponding decision rules of MS3WDA. The key steps and specific algorithm are summarized for MS3WDA. Finally, a case study is provided to demonstrate the practicability and feasibility of MS3WDA. Meanwhile, the rationality, robustness and superiority of MS3WDA are further validated by the sensitivity analysis and comparative analysis.

Imbalanced data classification is one of the challenging problems in machine learning. Oversampling is a promising technique that generates synthetic minority instances to balance the dataset. Inappropriate minority instances generated may deteriorate the performance of the classifier. Majority of the oversampling algorithms create new minority instances by choosing nearest neighbors for random interpolation. However, these methods do not provide new information to the dataset and therefore standard classifiers do not show good performance on such datasets. Therefore, it is necessary to generate diverse minority class instances to increase the performance of the classifier. Since, every feature of each minority class instance contribute valuable information, generating synthetic instances from the features of all minority instances would produce diverse minority instances, thereby increasing the performance of the classifier. This paper proposes a Hierarchical Heterogeneous Ant Colony Optimization based oversampling algorithm using Feature Similarity (HHACO-FSOTe) for generation of synthetic minority instances. Instead of choosing few neighbors for interpolation, the proposal considers all minority instances for generation of synthetic instances. HHACO-FSOTe generates new feature values by computing the minimum absolute difference between the features of a given minority instance and the corresponding features of the remaining minority instances. The features in the dataset are distributed among the ant agents enabling parallelism, thereby reducing the time taken for oversampling. HHACO-FSOTe do not require parameter tuning or training. The proposal is evaluated on 41 low dimensional, 11 high dimensional and 8 noisy datasets. Experiments reveal that HHACO-FSOTe is competent with the state-of-art oversampling techniques. Results were validated using non-parametric statistical tests.

In complex system, health state assessment can determine the state of the system and identify potential system problems. However, due to the numerous uncertainties and variations present in complex systems, it is difficult to effectively construct assessment models. Belief rule base (BRB) can use data-driven and knowledge-driven methods to effectively address uncertain information, and is widely used for modeling health state assessments of complex systems. The primary modeling and optimization goals of BRB is currently at accuracy, ignoring the impact of robustness on complex systems, and the reliability of the model is reduced. Therefore, this article introduces a novel method to balance the accuracy and robustness of BRB models. This method enhances the performance of the BRB model in assessing complex system health and provides valuable guidance for engineering applications. Firstly, the guidelines for BRB modeling are systematically summarized to address the trade-off between accuracy and robustness. This provides essential guidance for constructing BRB models during the model-building process. Secondly, four feasible domain criteria are proposed to enhance the reliability of the BRB during the model optimization process. A modified multi-objective optimization algorithm is proposed based on the feasible domain criteria. Finally, in the case studies of aerospace relay and lithium-ion battery health assessments, the MSE of the proposed model for aerospace relay health assessment is 0.0015 with a Lipschitz constant of 6.73, while for lithium-ion battery health assessment, the MSE is 0.0013 with a Lipschitz constant of 24.17. The experimental results demonstrate that the proposed model has an advantage in terms of the trade-offs between both robustness and accuracy.

Load frequency control (LFC) is a significant control problem in the operation of interconnected power systems. It keeps the change in system frequency within specific limits by maintaining the balance between power generation and load demand. In modern interconnected power systems, various control strategies, including conventional control techniques and other data-driven approaches, have been adopted to improve the effectiveness of LFC. The control technique based on reinforcement learning (RL) is one of the contemporary data-driven control strategies for LFC. Recently, the attention of researchers has surged towards RL-based control strategies for LFC. Several survey literature has been published in the field of LFC concerning the various control strategies for the effective operation of the power system. However, these surveys have not considered a complete systematic review of RL-driven LFC. An exhaustive review is essential to demonstrate the current status and identify future advancements in this field. This paper presents a comprehensive review of LFC based on the RL-driven control strategy. This study begins by presenting a mathematical and conceptual understanding of reinforcement learning. Finally, a broad classification of RL algorithms and the algorithm-wise literature survey on LFC are provided extensively. This comprehensive and insightful literature survey may serve as a valuable resource for the researchers, addressing the gaps between recent advances, implementation difficulties, and future developments in LFC using the RL-driven control strategy.

The emergence of high-quality deepfake facial videos has raised concerns about the security of facial images. Existing face forgery detectors mainly tend to locate a specific forgery region of the human face for detection, which achieves satisfactory performance with known forgery patterns presented in the training set. However, with the continuous advancements in face forgery technology, this approach becomes less reliable with new forgery patterns that emerge. Towards this end, we proposed a novel Self-supervised Face Geometry Information Analysis Network (SF-GAN) method for generalized face forgery detection. SF-GAN effectively leverages the relationships among informative regions based on information theory. Drawing from information theory, regions with high uncertainty tend to contain more valuable information. Our methodology integrates a self-supervised learning mechanism, enabling the precise identification of multiple informative regions. Furthermore, we leverage facial geometry by establishing both explicit and latent geometric relationships through the use of Graph Convolutional Networks (GCNs). Within our framework, facial landmarks and informative regions are depicted as nodes in the GCNs. By analyzing the geometric relationships between the graph of facial landmarks and the graph of informative regions, we are able to identify valid anomalous regions, thereby minimizing uncertainty. Our proposed model gains a comprehensive understanding of common information in face forgery images. Extensive experiments on eight large-scale benchmark datasets: FaceForensics++ (FF++), WildDeepfake (WDF), Celeb-DF v2 (CDF), DeepFake Detection Challenge (DFDC), DFDC preview (DFDC-P), Deepfake Detection (DFD), DeeperForensics-1.0 (DF-1.0) and ForgeryNIR, show that the proposed method is comparable to state-of-the-arts and exhibits better generalizability. Specifically, our SF-GAN, when trained on high-quality FF++ data, achieves an impressive AUC of 76.43% on the CDF dataset.