Decision Fusion Approach Research Articles

Wind Farm Prediction of Icing Based on SCADA Data

In cold climates, ice formation on wind turbines causes power reduction produced by a wind farm. This paper introduces a framework to predict icing at the farm level based on our recently developed Temporal Convolutional Network prediction model for a single turbine using SCADA data.First, a cross-validation study is carried out to evaluate the extent predictors trained on a single turbine of a wind farm can be used to predict icing on the other turbines of a wind farm. This fusion approach combines multiple turbines, thereby providing predictions at the wind farm level. This study shows that such a fusion approach improves prediction accuracy and decreases fluctuations across different prediction horizons when compared with single-turbine prediction. Two approaches are considered to conduct farm-level icing prediction: decision fusion and feature fusion. In decision fusion, icing prediction decisions from individual turbines are combined in a majority voting manner. In feature fusion, features of individual turbines are averaged first before conducting prediction. The results obtained indicate that both the decision fusion and feature fusion approaches generate farm-level icing prediction accuracies that are 7% higher with lower standard deviations or fluctuations across different prediction horizons when compared with predictions for a single turbine.

Fusing hyperspectral imaging and electronic nose data to predict moisture content in Penaeus vannamei during solar drying.

The control of moisture content (MC) is essential in the drying of shrimp, directly impacting its quality and shelf life. This study aimed to develop an accurate method for determining shrimp MC by integrating hyperspectral imaging (HSI) with electronic nose (E-nose) technology. We employed three different data fusion approaches: pixel-, feature-, and decision-fusion, to combine HSI and E nose data for the prediction of shrimp MC. We developed partial least squares regression (PLSR) models for each method and compared their performance in terms of prediction accuracy. The decision fusion approach outperformed the other methods, producing the highest determination coefficients for both calibration (0.9595) and validation sets (0.9448). Corresponding root-mean square errors were the lowest for the calibration set (0.0370) and validation set (0.0443), indicating high prediction precision. Additionally, this approach achieved a relative percent deviation of 3.94, the highest among the methods tested. The findings suggest that the decision fusion of HSI and E nose data through a PLSR model is an effective, accurate, and efficient method for evaluating shrimp MC. The demonstrated capability of this approach makes it a valuable tool for quality control and market monitoring of dried shrimp products.

Elasticity unleashed: Fine-grained cloud scaling through distributed three-way decision fusion with multi-head attention

With the proliferation of cloud computing, elastic resource scaling has emerged as a critical challenge. Cloud platforms must efficiently allocate resources to match dynamic workloads. This study addresses the key trade-off between responsiveness and prudence in scaling a novel distributed three-way decision fusion approach. Distributed three-way decisions of immediate, delayed, or no scaling are made at coarse and fine-time granularity levels. Subsequently, a multi-head attention mechanism intelligently fuses these decisions, weighing the relevance of the different granularities to synthesize an integrated scaling strategy. The reactiveness of sudden workload changes are balanced during fusion by considering long-term trends. Comprehensive experiments on real-world data demonstrated that the proposed fusion strategy substantially improves resource efficiency and adherence to service-level agreements compared to existing methods. Multi-head attention imparts autonomy in adapting to diverse operating conditions. The proposed principled methodology and attention-based decision aggregation hold significance for efficient, adaptive, and interpretable cloud scaling mechanisms.

Deep Learning-based NSCLC Classification from Whole-Slide Images: Leveraging Expectation-Maximization and InceptionV3

The standard-of-care treatment for Non-Small Cell Lung Cancer (NSCLC) is carried out via hematoxylin and eosin (H&E)-stained whole slide tissue images (WSI). These WSIs assist in classifying NSCLC into its prominent subtypes: adenocarcinoma (LUAD) and squamous cell carcinoma (LUSC)– a process, though efective, yet cumbersome for the pathologists when performed manually. To assist medical practitioners, a robust Artificial intelligence-based model (such as convolutional neural networks (CNN)) could be utilized to classify WSIs automatically. However, training gigapixel-level WSIs directly on a CNN is computationally prohibitive. This work proposes a novel EM-Inception model for efficient classification of NSCLC WSIs via a decision fusion approach. For effective memory utilization, the WSIs are initially segmented into patches. Next, a subset of discriminative patches (high likelihood of having tumorous region) is identifed with the assistance of the InceptionV3-based model and the EM algorithm–the model training is started with all the patches of each WSI, consecutively executing the EM algorithm to retain only the discriminative patches based on the patch-level predictions of the model. Finally, the WSI-level predictions are obtained by aggregating the discriminative patch-level predictions using the voting method. Post-training, EM-Inception’s classification efficacy is evaluated using 10-fold cross-validation. The proposed model achieves a classification performance of [0.87 ± 0.03] accuracy, [0.85 ± 0.12] specificity, [0.89±0.12] sensitivity, [0.87±0.03] balanced accuracy, [0.87±0.03] F1-score, and [0.917±0.03] AU-ROC score, each at 95% confidence level. The performance of the proposed model is also comparable to the state-of-the-art works. This illustrates the effectiveness of EM-Inception in lung cancer classification. As a future course of work, the eXplainable AI tools could be leveraged to provide deeper insights into the WSIs, such as highlighting the densely tumorous regions

Detecting the invasive Lespedeza cuneata in grasslands using commercial small satellite imagery

ABSTRACT Lespedeza cuneata (L. cuneata) is an invasive legume threatening grassland ecosystems of several states in the U.S. Mapping the occurrence of invasive species is the necessary first step to control their spread. Imaging spectroscopy or hyperspectral remote sensing imagery has shown its effectiveness in detecting invasive plants. However, challenges such as limited access to hyperspectral imagery, coarse temporal resolution, high costs, and lack of standardized analysis approaches have limited their application in monitoring biological invasions. Here, we aimed to take advantage of fine spatial and temporal resolution of PlanetScope multispectral satellite data and develop an effective approach to detect L. cuneata in a naturally assembled grassland ecosystem. To achieve this objective, we first sampled a suite of foliar functional traits in the field and identified key traits that distinguish L. cuneata from native plants. We then used our airborne hyperspectral data to select vegetation indices that are associated with the key traits while focusing on those indices that are applicable to satellite-derived multispectral data. We developed seven machine learning models and a decision fusion approach using layers of vegetation indices to map L. cuneata during the growing season. Using our approach, we were able to map L. cuneata with high level of accuracy using PlanetScope multispectral imagery. Our results suggested that the optimal period for detecting L. cuneata in our site was mid-to-late growing season, with an overall classification accuracy of 80.49 ± 5.69% (mid-August) and 82.02 ± 4.66% (end of September), respectively. Our study contributes to developing a cost-effective solution for mapping and monitoring invasive plants, which is essential for effective control of biological invasions.

DETECTION OF PRETERM BIRTH FROM THE NONCONTRACTION SEGMENTS OF UTERINE EMG USING HJORTH PARAMETERS AND SUPPORT VECTOR MACHINE

Uterine electromyography (uEMG) measures the electrical activity of the uterus noninvasively and is a promising technique for detecting preterm birth. Nevertheless, uterine contractions are irregular during pregnancy and may not present during standard 30-min recording. Hence, this study analyzes the noncontraction of uEMG signals for predicting premature birth. Three channels of 53 and 47 noncontraction segments under the term and preterm conditions, respectively, are obtained from the publicly available database. The signals are preprocessed, and the contractions and noncontraction segments are extracted manually based on the annotations. The Hjorth features, namely activity, mobility, and complexity, are extracted from the signals. Classification algorithms, namely support vector machine, random forest, and adaptive boosting classifier, are designed to distinguish between term and preterm conditions. The results show that mobility decreases, and complexity increases in preterm conditions. The support vector machine based on the proposed features of a single channel yields a maximum accuracy of 84.3% and F1-score of 82.8% in differentiating term and preterm conditions. In order to improve the performance further, we adapted a decision fusion approach that combines predictions from multiple channels. The improved model enhances the accuracy and F1-score by about 3%. Therefore, it appears that the proposed approach using noncontraction segments could be used as a biomarker for the reliable prediction of premature birth.

A trusted decision fusion approach for the power internet of things with federated learning

The power Internet of Things generates a large amount of data at any time, which can be transformed into precise decisions with the help of artificial intelligence approaches. However, the owners of electricity data with boundaries are often concerned with data leakage. Therefore, when building models that feed big data into deep learning artificial intelligence approaches for precise decision-making within the power Internet of Things, it is essential to ensure the data’s security. This paper proposes a framework for model training and decision making system applied to the field of power IoT, which consists of two parts: data security sharing and hierarchical decision making. The proposed framework utilizes a homomorphic encryption-based federated learning approach to protect private data from leakage. In addition, data augmentation and transfer learning are used to address the issue of insufficient local training data. Moreover, the framework attempts to incorporate the specialized nature of traditional manual decision-making in the power field by fusing expert and model values after stratifying the requirements. Experiments are conducted to simulate the decision requirements in the field of power Internet of Things (e.g., electrical material identification), using image recognition as an example. The experimental results show that the proposed models can achieve high accuracy rates and the fusion approach is feasible.

Ensemble Deep Learning Ultimate Tensile Strength Classification Model for Weld Seam of Asymmetric Friction Stir Welding

Friction stir welding is a material processing technique used to combine dissimilar and similar materials. Ultimate tensile strength (UTS) is one of the most common objectives of welding, especially friction stir welding (FSW). Typically, destructive testing is utilized to measure the UTS of a welded seam. Testing for the UTS of a weld seam typically involves cutting the specimen and utilizing a machine capable of testing for UTS. In this study, an ensemble deep learning model was developed to classify the UTS of the FSW weld seam. Consequently, the model could classify the quality of the weld seam in relation to its UTS using only an image of the weld seam. Five distinct convolutional neural networks (CNNs) were employed to form the heterogeneous ensemble deep learning model in the proposed model. In addition, image segmentation, image augmentation, and an efficient decision fusion approach were implemented in the proposed model. To test the model, 1664 pictures of weld seams were created and tested using the model. The weld seam UTS quality was divided into three categories: below 70% (low quality), 70–85% (moderate quality), and above 85% (high quality) of the base material. AA5083 and AA5061 were the base materials used for this study. The computational results demonstrate that the accuracy of the suggested model is 96.23%, which is 0.35% to 8.91% greater than the accuracy of the literature’s most advanced CNN model.

Performance Evaluation of Different Decision Fusion Approaches for Image Classification

Image classification is one of the major data mining tasks in smart city applications. However, deploying classification models that have good generalization accuracy is highly crucial for reliable decision-making in such applications. One of the ways to achieve good generalization accuracy is through the use of multiple classifiers and the fusion of their decisions. This approach is known as “decision fusion”. The requirement for achieving good results with decision fusion is that there should be dissimilarity between the outputs of the classifiers. This paper proposes and evaluates two ways of attaining the aforementioned dissimilarity. One is using dissimilar classifiers with different architectures, and the other is using similar classifiers with similar architectures but trained with different batch sizes. The paper also compares a number of decision fusion strategies.

Decision fusion approach for detecting unknown wafer bin map patterns based on a deep multitask learning model

In semiconductor manufacturing, wafer bin maps (WBMs) present specific defect patterns that provide crucial information for tracking abnormal processes. Thus, it is necessary to correctly classify WBM defects to achieve yield improvements. However, unknown types of defects constantly emerge due to the development of new process technologies and devices, degrading classification performance. Thus, a novel open set recognition (OSR) method that aims to detect unknown defects while correctly classifying known defect types is proposed in this work. The proposed method first extracts the reconstruction errors using an autoencoder and the random network errors using a random network distillation technique. Next, the two types of errors are combined by applying the extreme value theory-based Weibull calibration technique to produce probability scores for unknown detection. Finally, a deep convolutional neural network classifier assigns known classes to the samples determined as known. In the proposed method, multiple networks are interconnected, and we apply a simple sequential multitask learning mechanism to coordinate all networks. Extensive experiments were conducted on a real-world WBM defect dataset, and the results showed that each component proposed in this paper helped improve the unknown detection performance of our approach while minimizing its closed set classification performance loss. As a result, the proposed method outperformed the state-of-the-art OSR methods, achieving area under the receiver operating characteristic curve (AUROC) increases of at least 0.124 for unknown detection cases.

Multi-Module Decision Fusion in Operational Status Monitoring

Multimodule and multisensor data with different characteristics can reflect the overall operating status of the equipment from different angles. It is far from enough to monitor the operating status of equipment from a single perspective for this fails to take all valid information into account. However, data integration may face problems such as the curse of dimensionality and scale mismatches. Therefore, decision fusion, which needs to measure and manage evidence conflicts, has attracted extensive attention from scholars. However, most of the state-of-art methods focus on conflict management based on evidence itself and ignore the irrationality of conflict factor <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$K$ </tex-math></inline-formula> in measuring conflicts and the reliability of evidence sources in conflict management. In order to solve the above problems, a novel hybrid decision fusion approach is proposed in this article. First, divide data into modules and use the models in the model library to conduct cross-validation, thus obtaining the performance ranking. Then, select the optimal classifier of each module to obtain the evidence for decision fusion. Given the conflict of evidences, the Jensen–Shannon (JS) divergence is used to measure the conflict, and those high-conflict evidences will be revised through sensitivity and support analyses. Finally, the Dempster–Shafer (DS) evidence theory is used to integrate multimodule evidences to assess the status. To prove the feasibility and effectiveness of this approach, a realistic operational shield case in China is used.

Decision fusion for reliable fault classification in energy-intensive process industries

Heavy industries such as oil & gas, steel & iron production, the cement industry, chemical processes, and pulp & paper mills are among the industries that contribute most to greenhouse gas emissions. The operation of processes in these industries requires massive amounts of energy, which is mainly generated through the consumption of natural resources, leading to the emission of harmful gases. These emissions can be reduced by improving the efficiency of process operations and with better management of abnormal events/faults through accurate fault classification. Managers and operators of these industries need intelligent fault classification strategies that provide accurate information and decisions on faulty situations management, thus helping day-to-day processes operate in safe, reliable and energy-efficient ways. This paper proposes a decision fusion approach that combines outputs produced by diversified machine learning fault classifiers, each with distinct pattern representations. The proposed approach mainly adopts the behavior knowledge space (BKS) method and updates the corresponding lookup table based on the F1-scores calculated for every single fault classifier based on historical data. The concept is based on the exploitation of the advantages of each fault classifier as a complementary information source to effectively classify faults and provide plant operators and engineers with unified, accurate and comprehensive decisions. The proposed approach is validated through two case studies in the pulp and paper industry. The first one is the pulp mill process benchmark using simulated data and the second case is on a reboiler in a Canadian thermomechanical pulp mill. The results obtained demonstrate that the accuracy of the proposed approach is higher than the accuracy of every single classifier and other comparable methods applied to such complex industrial processes. The results have helped the mill operators correctly identify the causes of abnormal events and have contributed to significant energy savings and reduction in Greenhouse Gas (GHG) emissions.

A novel multi-information decision fusion based on improved random forests in HVCB fault detection application

Evaluating the mechanical state of high-voltage circuit breakers (HVCBs) based on vibration information has currently become an important research direction. In contrast to the unicity of the travel–time and current–time curves, the vibration information from the different positions is diverse. These differences are often overlooked in HVCB fault identification applications. Additionally, the fault recognition results based on different location information often vary, and conflicting diagnosis results directly cause the accurate identification of the fault type to fail. Therefore, in this paper, a novel multi-information decision fusion approach is proposed based on the improved random forest (RF) and Dempster-Shafer evidence theory. In the proposed method, the diagnostic distribution of all classification regression trees in the RF is considered to solve the conflicts among the multi-information diagnosis results. Experimental results show that the proposed method eases the contradiction of multi-position diagnostic results and improves the accuracy of fault identification. Furthermore, compared to the common classifiers and probability generation methods, the effectiveness and superiority of the proposed method are verified.

Decision Fusion Approach for Detecting Unknown Wafer Bin Map Patterns Based on a Deep Multitask Learning Model

Decision Fusion Approach for Detecting Unknown Wafer Bin Map Patterns Based on a Deep Multitask Learning Model

Novel Multiscale Decision Fusion Approach to Unsupervised Change Detection for High-Resolution Images

High-resolution remote sensing images usually contain multiscale information, which can be used to enhance the change detection (CD) performance. How to make effective use of multiscale information needs intensive study. This letter presents a novel multiscale decision fusion (MDF) method for unsupervised CD based on Dempster–Shafer (DS) theory and modified conditional random field (CRF). The method consists of three main steps: 1) images of three different scales are created automatically by image segmentation, and then three-scale difference images (DIs) are produced by applying change vector analysis to the three-scale images; 2) the membership function of each scale DI is estimated by fuzzy clustering, and the fusion membership, as well as an initial CD map, is obtained by combining the estimated membership using DS theory; and 3) the initial CD map is refined with an improved CRF that incorporates a spatial attraction model. The proposed method can combine the multiscale information in images and the spatial contextual information. The effectiveness of the proposed method was validated by two experiments with high-resolution remote sensing images.

Multi-segment Majority Voting Decision Fusion for MI EEG Brain-Computer Interfacing

Brain-computer interfaces (BCIs) based on the electroencephalogram (EEG) generated during motor imagery (MI) have the potential to be used in brain-controlled prosthetics, neurorehabilitation and gaming. Many MI EEG classification systems segment EEG into windows for classification. However, a comprehensive analysis of decision fusion based on the segmented EEG data, within the context of different classifiers, has not been carried out. This study presents a multi-segment majority voting (MSMV) decision fusion approach in which an EEG trial is segmented using overlapping windows. Segments are labelled and a final classification label for the trial is derived through majority voting, using the common spatial pattern (CSP) features. The impact of the MSMV approach on the classification accuracy of six classifiers was investigated. The effects of window size and overlap were analysed. Results were generated using five different subsets of EEG channels, and channel subsets for static EEG analysis are also proposed. The BCI Competition III dataset IVa was used. The MSMV decision fusion approach was found to significantly improve the classification accuracy for linear discriminant analysis (LDA), support vector machine (SVM), naïve-Bayes (NB) and random forest (RF) classifiers. The classification accuracy was improved by 5.02%, 4.41%, 1.25% and 3.62% for the SVM, LDA, NB and RF classifiers, respectively. The channel analysis indicated the importance of central-parietal and central-frontal electrode regions for MI EEG classification. MSMV decision fusion improved MI EEG classification performance and could be considered for future studies, particularly in online systems that deal with buffered data.

An Evaluation of Traditional and CNN-Based Feature Descriptors for Cartoon Pornography Detection

Inappropriate visual content on the internet has spread everywhere, and thus children are exposed unintentionally to sexually explicit visual content. Animated cartoon movies sometimes have sensitive content such as pornography and sex. Usually, video sharing platforms take children’s e-safety into consideration through manual censorship, which is both time-consuming and expensive. Therefore, automated cartoon censorship is highly recommended to be integrated into media platforms. In this paper, various methods and approaches were explored to detect inappropriate visual content in cartoon animation. First, state-of-the-art conventional feature techniques were utilised and evaluated. In addition, a simple end-to-end convolutional neural network (CNN) was used and was found to outperform conventional techniques in terms of accuracy (85.33%) and F1 score (83.46%). Additionally, to target the deeper version of CNNs, ResNet, and EfficientNet were demonstrated and compared. The CNN-based extracted features were mapped into two classes: normal and porn. To improve the model’s performance, we utilised feature and decision fusion approaches which were found to outperform state-of-the-art techniques in terms of accuracy (87.87%), F1 score (87.87%), and AUC (94.40%). To validate the domain generalisation performance of the proposed methods, CNNs, pre-trained on the cartoon dataset were evaluated on public NPDI-800 natural videos and found to provide an accuracy of 79.92%, and F1 score of 80.58%. Similarly, CNNs, pre-trained on the public NPDI-800 natural videos, were evaluated on cartoon dataset and found to give an accuracy of 82.666%, and F1 score of 81.588%. Finally, a novel cartoon pornography dataset, with various characters, skin colours, positions, viewpoints, and scales, was proposed.

Adjustive decision fusion approaches for hyperspectral image classification

Adjustive decision fusion approaches for hyperspectral image classification

Emotion Assessment Using Feature Fusion and Decision Fusion Classification Based on Physiological Data: Are We There Yet?

Emotion recognition based on physiological data classification has been a topic of increasingly growing interest for more than a decade. However, there is a lack of systematic analysis in literature regarding the selection of classifiers to use, sensor modalities, features and range of expected accuracy, just to name a few limitations. In this work, we evaluate emotion in terms of low/high arousal and valence classification through Supervised Learning (SL), Decision Fusion (DF) and Feature Fusion (FF) techniques using multimodal physiological data, namely, Electrocardiography (ECG), Electrodermal Activity (EDA), Respiration (RESP), or Blood Volume Pulse (BVP). The main contribution of our work is a systematic study across five public datasets commonly used in the Emotion Recognition (ER) state-of-the-art, namely: (1) Classification performance analysis of ER benchmarking datasets in the arousal/valence space; (2) Summarising the ranges of the classification accuracy reported across the existing literature; (3) Characterising the results for diverse classifiers, sensor modalities and feature set combinations for ER using accuracy and F1-score; (4) Exploration of an extended feature set for each modality; (5) Systematic analysis of multimodal classification in DF and FF approaches. The experimental results showed that FF is the most competitive technique in terms of classification accuracy and computational complexity. We obtain superior or comparable results to those reported in the state-of-the-art for the selected datasets.

A Quantum-Like multimodal network framework for modeling interaction dynamics in multiparty conversational sentiment analysis

Sentiment analysis in conversations is an emerging yet challenging artificial intelligence (AI) task. It aims to discover the affective states and emotional changes of speakers involved in a conversation on the basis of their opinions, which are carried by different modalities of information (e.g., a video associated with a transcript). There exists a wealth of intra- and inter-utterance interaction information that affects the emotions of speakers in a complex and dynamic way. How to accurately and comprehensively model complicated interactions is the key problem of the field. To fill this gap, in this paper, we propose a novel and comprehensive framework for multimodal sentiment analysis in conversations, called a quantum-like multimodal network (QMN), which leverages the mathematical formalism of quantum theory (QT) and a long short-term memory (LSTM) network. Specifically, the QMN framework consists of a multimodal decision fusion approach inspired by quantum interference theory to capture the interactions within each utterance (i.e., the correlations between different modalities) and a strong-weak influence model inspired by quantum measurement theory to model the interactions between adjacent utterances (i.e., how one speaker influences another). Extensive experiments are conducted on two widely used conversational sentiment datasets: the MELD and IEMOCAP datasets. The experimental results show that our approach significantly outperforms a wide range of baselines and state-of-the-art models.