Feature-level Fusion Strategy Research Articles

Deep learning based crop-type mapping using SAR and optical data fusion

Accurate crop maps are essential in various applications related to food security management activities. Remote Sensing is the primary data source for land-use and land-cover monitoring applications. However, high spectral, spatial, and temporal variations of crop types during their phonological stages make crop mapping challenging. According to the complementary behaviors of optical and radar data, a deep learning-based feature-level fusion strategy is proposed for accurate crop-type mapping. The proposed method uses Sparse Auto-Encoders (SAE) as its base classifier and can involve spatial information in two data-preparation and post-processing steps. Involving spatial information in the classification process can significantly improve crop mapping. The post-processing step uses the guided filter as an edge-aware filtering process that uses field boundary information that implicitly exists in data. Accuracy analysis was performed using two pairs of RapidEye and UAVSAR data from two agricultural areas in Canada. The suitability and higher performance of the proposed method were demonstrated compared to traditional machine learning and common decision-level fusion methods. Accuracy metrics were improved by at least 3% and 1% in two datasets. Class F-scores were also significantly enhanced in oat and wheat classes (up to 20%).

Non-destructive discrimination of honey origin based on multispectral information fusion technology

Accurate discrimination of honey origin is of significant importance for safeguarding consumer rights and promoting the sustainable development of apiculture. Spectroscopic techniques, as rapid, efficient, and non-destructive detection methods, have been widely applied in honey research. However, studies often focus on a single analytical technique, limiting the acquisition of comprehensive chemical information and hindering improvements in discrimination accuracy. Data fusion, an information processing technique that enhances test results' accuracy by integrating data from multiple devices, holds vast potential for various applications. Therefore, this study employed data fusion technology to investigate 142 honey samples from five different origins by combining Fourier Transform Mid-Infrared Spectroscopy (FT-MIR) and Raman spectroscopy. Discrimination models based on single-spectral and multispectral Probabilistic Neural Networks (PNN) and Support Vector Machines (SVM) were established using data-level and feature-level fusion strategies. Compared to single-spectral methods, multispectral data fusion significantly improved the geospatial traceability discrimination performance of the models. Particularly, the feature-level fusion strategy demonstrated significant discrimination effectiveness for honey origin. By optimizing the optimal SVM model with the Particle Swarm Optimization (PSO) algorithm, the final test accuracy reached 95.28 %, representing an 11.4 % increase over the best single-spectrum model's test accuracy. Therefore, the establishment of a PSO-SVM discrimination model using feature-level fusion technology enables rapid discrimination of honey origin, providing a reliable and stable method for non-destructive honey origin discrimination.

A novel feature-level fusion scheme with multimodal attention CNN for heart sound classification

Background and objectiveMost of the existing machine learning-based heart sound classification methods achieve limited accuracy. Since they primarily depend on single domain feature information and tend to focus equally on each part of the signal rather than employing a selective attention mechanism. In addition, they fail to exploit convolutional neural network (CNN) - based features with an effective fusion strategy. MethodsIn order to overcome these limitations, a novel multimodal attention convolutional neural network (MACNN) with a feature-level fusion strategy, in which Mel-cepstral domain as well as general frequency domain features are incorporated to increase the diversity of the features, is proposed in this paper. In the proposed method, DilationAttenNet is first utilized to construct attention-based CNN feature extractors and then these feature extractors are jointly optimized in MACNN at the feature-level. The attention mechanism aims to suppress irrelevant information and focus on crucial diverse features extracted from the CNN. ResultsExtensive experiments are carried out to study the efficacy of the feature level fusion in comparison to that with early fusion. The results show that the proposed MACNN method significantly outperforms the state-of-the-art approaches in terms of accuracy and score for the two publicly available Github and Physionet datasets. ConclusionThe findings of our experiments demonstrated the high performance for heart sound classification based on the proposed MACNN, and hence have potential clinical usefulness in the identification of heart diseases. This technique can assist cardiologists and researchers in the design and development of heart sound classification methods.

TAG-Net: Target Attitude Angle-Guided Network for Ship Detection and Classification in SAR Images

Synthetic aperture radar (SAR) ship detection and classification has gained unprecedented attention due to its important role in maritime transportation. Many deep learning-based detectors and classifiers have been successfully applied and achieved great progress. However, ships in SAR images present discrete and multi-centric features, and their scattering characteristics and edge information are sensitive to variations in target attitude angles (TAAs). These factors pose challenges for existing methods to obtain satisfactory results. To address these challenges, a novel target attitude angle-guided network (TAG-Net) is proposed in this article. The core idea of TAG-Net is to leverage TAA information as guidance and use an adaptive feature-level fusion strategy to dynamically learn more representative features that can handle the target imaging diversity caused by TAA. This is achieved through a TAA-aware feature modulation (TAFM) module. It uses the TAA information and foreground information as prior knowledge and establishes the relationship between the ship scattering characteristics and TAA information. This enables a reduction in the intra-class variability and highlights ship targets. Additionally, considering the different requirements of the detection and classification tasks for the scattering information, we propose a layer-wise attention-based task decoupling detection head (LATD). Unlike general deep learning methods that use shared features for both detection and classification tasks, LATD extracts multi-level features and uses layer attention to achieve feature decoupling and select the most suitable features for each task. Finally, we introduce a novel salient-enhanced feature balance module (SFB) to provide richer semantic information and capture the global context to highlight ships in complex scenes, effectively reducing the impact of background noise. A large-scale ship detection dataset (LSSDD+) is used to verify the effectiveness of TAG-Net, and our method achieves state-of-the-art performance.

Applying Hybrid Clustering in Pulsar Candidate Sifting with Multi-modality for FAST Survey

Pulsar search is always the basis of pulsar navigation, gravitational wave detection and other research topics. Currently, the volume of pulsar candidates collected by the Five-hundred-meter Aperture Spherical radio Telescope (FAST) shows an explosive growth rate that has brought challenges for its pulsar candidate filtering system. Particularly, the multi-view heterogeneous data and class imbalance between true pulsars and non-pulsar candidates have negative effects on traditional single-modal supervised classification methods. In this study, a multi-modal and semi-supervised learning based on a pulsar candidate sifting algorithm is presented, which adopts a hybrid ensemble clustering scheme of density-based and partition-based methods combined with a feature-level fusion strategy for input data and a data partition strategy for parallelization. Experiments on both High Time Resolution Universe Survey II (HTRU2) and actual FAST observation data demonstrate that the proposed algorithm could excellently identify pulsars: On HTRU2, the precision and recall rates of its parallel mode reach 0.981 and 0.988 respectively. On FAST data, those of its parallel mode reach 0.891 and 0.961, meanwhile, the running time also significantly decreases with the increment of parallel nodes within limits. Thus, we can conclude that our algorithm could be a feasible idea for large scale pulsar candidate sifting for FAST drift scan observation.

A Novel Multi-Modal Feature-Level Fusion Scheme for High Accurate Indoor Localization

A Novel Multi-Modal Feature-Level Fusion Scheme for High Accurate Indoor Localization

MASKED AND UNMASKED FACE RECOGNITION ON UNCONSTRAINED FACIAL IMAGES USING HAND-CRAFTED METHODS

In this study, the face recognition task is applied on masked and unmasked faces using hand-crafted methods. Due to COVID-19 and masks, facial identification from unconstrained images became a hot topic. To avoid COVID-19, most people use masks outside. In many cases, typical facial recognition technology is useless. The majority of contemporary advanced face recognition methods are based on deep learning, which primarily relies on a huge number of training examples, however, masked face recognition may be investigated using hand-crafted approaches at a lower computing cost than using deep learning systems. A low-cost system is intended to be constructed for recognizing masked faces and compares its performance to that of face recognition systems that do not use masks. The proposed method fuses hand-crafted methods using feature-level fusion strategy. This study compares the performance of masked and unmasked face recognition systems. The experiments are undertaken on two publicly accessible datasets for masked face recognition: Masked Labeled Faces in the Wild (MLFW) and Cross-Age Labeled Faces in the Wild (CALFW). The best accuracy is achieved as 94.8% on MLFW dataset. The rest of the results on different train and test sets from CALFW and MLFW datasets are encouraging compared to the state-of-the-art models.

A Comparison of Different Data Fusion Strategies’ Effects on Maize Leaf Area Index Prediction Using Multisource Data from Unmanned Aerial Vehicles (UAVs)

The leaf area index (LAI) is an important indicator for crop growth monitoring. This study aims to analyze the effects of different data fusion strategies on the performance of LAI prediction models, using multisource images from unmanned aerial vehicles (UAVs). For this purpose, maize field experiments were conducted to obtain plants with different growth status. LAI and corresponding multispectral (MS) and RGB images were collected at different maize growth stages. Based on these data, different model design scenarios, including single-source image scenarios, pixel-level multisource data fusion scenarios, and feature-level multisource data fusion scenarios, were created. Then, stepwise multiple linear regression (SMLR) was used to design LAI prediction models. The performance of models were compared and the results showed that (i) combining spectral and texture features to predict LAI performs better than using only spectral or texture information; (ii) compared with using single-source images, using a multisource data fusion strategy can improve the performance of the model to predict LAI; and (iii) among the different multisource data fusion strategies, the feature-level data fusion strategy performed better than the pixel-level fusion strategy in the LAI prediction models. Thus, a feature-level data fusion strategy is recommended for the creation of maize LAI prediction models using multisource UAV images.

Bolt tightness monitoring using multiple reconstructed narrowband Lamb waves combined with piezoelectric ultrasonic transducer

Narrowband tone burst excitation with a carefully selected frequency is generally utilized to generate Lamb waves for mode purity, dispersion diminution and signal interpretability. Narrowband excitation exhibits saturation phenomenon and even non-monotonic trends with the development of bolt looseness. In this research, a piezoelectric ultrasonic transducer with high signal-to-noise ratio is designed and fabricated for bolt tightness monitoring. The chirp signal is utilized to drive the actuator to generate broadband Lamb waves in the connected plate, and a transfer function-based signal reconstruction algorithm is innovatively proposed to extract narrowband tone burst responses with different center frequencies from the broadband chirp response for bolt tightness monitoring. The leaked wave energy-based bolt tightness indexes (TIs) are calculated based on the extracted multiple tone burst responses and a feature-level data fusion strategy is proposed to combine the TIs for exploiting the merits of different-frequency inspecting waves for different bolt tightness conditions. The fused TI not only presents a monotonic tendency with the increasing of bolt tightness, but also tackles the low sensitivity of narrowband Lamb waves for embryo bolt looseness. The proposed method contributes a novel and stable scheme for bolt tightness monitoring and opens a new perspective for damage evaluation of structures.

An Alzheimer’s disease classification method using fusion of features from brain Magnetic Resonance Image transforms and deep convolutional networks

Alzheimer’s is a progressive and irreversible brain degenerative disorder, and presenting an accurate early-stage diagnosis tool is vital for preventing disease progression. The previous research considered the features only from the brain Magnetic Resonance Image (MRI) or its transforms. This study presents two efficient fusion schemes to combine the deep features obtained from the brain MR image and its transforms. We consider the Orthogonal Ripplet II Transform (ORT-II), the Two-Dimensional Discrete Orthonormal Stockwell Transform (2D DOST), and the original brain MR image. We present the decision- and feature-level fusion schemes to combine the information of different inputs. In decision-level fusion, Convolutional Neural Networks (CNNs) separately classify each matrix, where all CNNs have the same structure, and the majority role makes the final decision. In feature-level fusion, the same CNNs extract each matrix’s features and then concatenate them to construct the feature vector. Since this vector contains many features, we employ Neighborhood Component Analysis (NCA) to select the more relevant features for classification to determine the level of dementia. We consider the different CNN models and classifiers to classify the brain MR images in three two-class scenarios. The results demonstrate that EfficientNet-B7 with an Artificial Neural Network (ANN) provides the highest accuracy. Also, feature-level fusion reaches a higher accuracy compared to the decision-level one.

Exploring fusion strategies for accurate RGBT visual object tracking

We address the problem of multi-modal object tracking in video and explore various options available for fusing the complementary information conveyed by the visible (RGB) and thermal infrared (TIR) modalities, including pixel-level, feature-level and decision-level fusion. Specifically, in contrast to the existing approaches, we propose and develop the paradigm for combining multi-modal information for image fusion at pixel level. At the feature level, two different kinds of fusion strategies are investigated for completeness, i.e., the attention-based online fusion strategy and the offline-trained fusion block. At the decision level, a novel fusion strategy is put forward, inspired by the success of the simple averaging configuration which has shown so much promise. The effectiveness of the proposed decision-level fusion strategy owes to a number of innovative contributions, including a dynamic weighting of the RGB and TIR contributions and a linear template update operation. A variant of the proposed decision fusion method produced the winning tracker at the Visual Object Tracking Challenge 2020 (VOT-RGBT2020). A comprehensive comparison of the innovative pixel and feature-level fusion strategies with the proposed decision-level fusion method highlights the advantages fusing multimodal information at the decision score level. Extensive experimental results on five challenging datasets, i.e., GTOT, VOT-RGBT2019, RGBT234, LasHeR and VOT-RGBT2020, demonstrate the effectiveness and robustness of the proposed method, compared to the state-of-the-art approaches. The Code is available at https://github.com/Zhangyong-Tang/DFAT.

ITCep: a deep learning framework for identification of T cell epitopes by harnessing fusion features.

Neoantigens recognized by cytotoxic T cells are effective targets for tumor-specific immune responses for personalized cancer immunotherapy. Quite a few neoantigen identification pipelines and computational strategies have been developed to improve the accuracy of the peptide selection process. However, these methods mainly consider the neoantigen end and ignore the interaction between peptide-TCR and the preference of each residue in TCRs, resulting in the filtered peptides often fail to truly elicit an immune response. Here, we propose a novel encoding approach for peptide-TCR representation. Subsequently, a deep learning framework, namely iTCep, was developed to predict the interactions between peptides and TCRs using fusion features derived from a feature-level fusion strategy. The iTCep achieved high predictive performance with AUC up to 0.96 on the testing dataset and above 0.86 on independent datasets, presenting better prediction performance compared with other predictors. Our results provided strong evidence that model iTCep can be a reliable and robust method for predicting TCR binding specificities of given antigen peptides. One can access the iTCep through a user-friendly web server at http://biostatistics.online/iTCep/, which supports prediction modes of peptide-TCR pairs and peptide-only. A stand-alone software program for T cell epitope prediction is also available for convenient installing at https://github.com/kbvstmd/iTCep/.

Multi-Biometrics: Survey and Projection of a New Biometric System

Multi-biometric systems using feature-level fusion allow more accuracy and reliability in recognition performance than uni-biometric systems. But in practice, this type of fusion is difficult to implement especially when we are facing heterogeneous biometric modalities or incompatible features. The major challenge of feature fusion is to produce a representation of each modality with an excellent level of discrimination. Beyond pure biometric modalities, the use of metadata has proven to improve the performance of biometric systems. In view of these findings, our work focuses on multi-origin biometrics which allows the use of pure biometric modalities and metadata in a feature fusion strategy. The main objective of this paper is to present an overview of biometrics as bordered in the literature with a particular focus on multibiometrics and to propose a model of a multi-origin biometric system using pure biometric and soft biometric modalities in a feature-level fusion strategy. The curvelet transformation and the order statistics are proposed respectively for the extraction the feature of the pure biometric modalities, and for the selection of the relevant feature of each modality in order to ensure a good level of discrimination of the individuals. In this paper, we have presented the overview of biometrics through its concepts, modalities, advantages, disadvantages and implementation architectures. A focus has been put on multi-biometrics with the presentation of a harmonized process for feature fusion. For the experiments, we proposed a global model for feature fusion in a multi-origin system using face and iris modalities as pure biometrics, and facial skin color as metadata. This system and the results will be presented in future work.

Monitoring black tea fermentation quality by intelligent sensors: Comparison of image, e-nose and data fusion

To scientifically and objectively monitor the fermentation quality of black tea, a computer vision system (CVS) and electronic nose (e-nose) were employed to analyze the black tea image and odor eigenvalues of Yinghong No. 9 black tea. First, the variation trends of tea polyphenols, volatile substances, image eigenvalues and odor eigenvalues with the extension of fermentation time were analyzed, and the fermentation process was categorized into three stages for classification. Second, principal component analysis (PCA) was employed on the image and odor eigenvalues obtained by CVS and e-nose. Partial least squares discriminant analysis (PLS-DA) was performed on 117 volatile components, and 51 differential volatiles were screened out based on variable importance in projection (VIP ≥1) and one-way analysis of variance (P < 0.05), including geraniol, linalool, nerolidol, and α-ionone. Then, image features and odor features are fused by using a data fusion strategy. Finally, the image, smell and fusion information were combined with random forest (RF), K-nearest neighbor (KNN) and support vector machine (SVM) to establish the classification models of different fermentation stages and to compare them. The results show that the feature-level fusion strategy integrating the SVM was the most efficient approach, with classification accuracy rates of 100% for the training sets and 95.6% for the testing sets. The performance of Support Vector Regression (SVR) prediction models for tea polyphenol content based on feature-level fusion data outperformed data-level models (Rc, RMSEC, Rp and RMSEP of 0.96, 0.48 mg/g, 0.94, 0.6 mg/g).

Towards enhancing emotion recognition via multimodal framework

Emotional AI is the next era of AI to play a major role in various fields such as entertainment, health care, self-paced online education, etc., considering clues from multiple sources. In this work, we propose a multimodal emotion recognition system extracting information from speech, motion capture, and text data. The main aim of this research is to improve the unimodal architectures to outperform the state-of-the-arts and combine them together to build a robust multi-modal fusion architecture. We developed 1D and 2D CNN-LSTM time-distributed models for speech, a hybrid CNN-LSTM model for motion capture data, and a BERT-based model for text data to achieve state-of-the-art results, and attempted both concatenation-based decision-level fusion and Deep CCA-based feature-level fusion schemes. The proposed speech and mocap models achieve emotion recognition accuracies of 65.08% and 67.51%, respectively, and the BERT-based text model achieves an accuracy of 72.60%. The decision-level fusion approach significantly improves the accuracy of detecting emotions on the IEMOCAP and MELD datasets. This approach achieves 80.20% accuracy on IEMOCAP which is 8.61% higher than the state-of-the-art methods, and 63.52% and 61.65% in 5-class and 7-class classification on the MELD dataset which are higher than the state-of-the-arts.

Suspicious Object Detection for Millimeter-Wave Images with Multi-View Fusion Siamese Network.

Millimeter-wave (MMW) imaging techniques have been widely used in the public security industries for their under-controlled privacy concerns and no health hazards. However, since MMW images are low resolution and most objects are small, reflection-weak, diverse, suspicious object detection in the MMW images is a very challenging task. This paper develops a robust suspicious object detector for the MMW images based on the Siamese network integrated with the pose estimation and image segmentation, which estimates the coordinates of human joints and segments the complete human images into symmetrical body part images. Unlike most existing detectors, which detect and recognize suspicious objects in MMW images and require a complete training set with correct annotations, our proposed model aims to learn the similarity between two symmetrical human body part images segmented from the complete MMW images. Furthermore, to decrease the misdetection caused by the restricted field of view, we further fuse the multi-view MMW images observed from the same person by designing a decision-level fusion strategy and feature-level fusion strategy based on the attention mechanism. Experimental results on the measured MMW images show that our proposed models have favorable detection accuracy and speed in practical application and thus prove their effectiveness.

Sea ice image classification based on ResFPG network and heterogeneous data fusion

ABSTRACT Sea ice detection has played an important role in climate protection and strategic deployment. Due to the autonomous learning characteristics of deep learning, it has been gradually applied to the classification of remote sensing sea ice images. At present, deep learning models are mostly used in sea ice classification of single source remote sensing data. Due to the limitations of single source data and the information loss of deep learning model in the process of feature extraction layer by layer, it is inevitable to encounter a bottleneck in sea ice detection requiring fine classification. To solve the above problems, this paper proposes a sea ice image classification method based on ResNet16-feature pyramid networks-spatial pyramid pooling-gated fusion network (ResFPG) and heterogeneous data fusion. In the feature extraction part, the method uses the improved ResNet16 to extract the multi-level feature information of sea ice in synthetic aperture radar data and optical data, reducing information loss in the feature extraction process, then mines and fuses the low-level spatial information and high-level semantic information through the improved feature pyramid networks (FPN), and then collects and fuses the output features of different scales through the spatial pyramid pooling (SPP) network. In the feature fusion part, a gated feature-level fusion strategy is designed to further improve the overall classification accuracy by adaptively adjusting the feature contribution of two heterogeneous sources of data through the gated fusion network (GFN). In order to verify the effectiveness of this method, we use two sets of heterogeneous sea ice remote sensing data located in Hudson Bay area for experiments. The experimental results show that compared with other image classification methods, the proposed method fully excavates and integrates the multi-scale and multi-level features in heterogeneous data, effectively distinguishes the feature contribution, and achieves better classification results (97.14% and 95.85%).

Multimodal deep learning model on interim [18F]FDG PET/CT for predicting primary treatment failure in diffuse large B-cell lymphoma.

The prediction of primary treatment failure (PTF) is necessary for patients with diffuse large B-cell lymphoma (DLBCL) since it serves as a prominent means for improving front-line outcomes. Using interim 18F-fluoro-2-deoxyglucose ([18F]FDG) positron emission tomography/computed tomography (PET/CT) imaging data, we aimed to construct multimodal deep learning (MDL) models to predict possible PTF in low-risk DLBCL. Initially, 205 DLBCL patients undergoing interim [18F]FDG PET/CT scans and the front-line standard of care were included in the primary dataset for model development. Then, 44 other patients were included in the external dataset for generalization evaluation. Based on the powerful backbone of the Conv-LSTM network, we incorporated five different multimodal fusion strategies (pixel intermixing, separate channel, separate branch, quantitative weighting, and hybrid learning) to make full use of PET/CT features and built five corresponding MDL models. Moreover, we found the best model, that is, the hybrid learning model, and optimized it by integrating the contrastive training objective to further improve its prediction performance. The final model with contrastive objective optimization, named the contrastive hybrid learning model, performed best, with an accuracy of 91.22% and an area under the receiver operating characteristic curve (AUC) of 0.926, in the primary dataset. In the external dataset, its accuracy and AUC remained at 88.64% and 0.925, respectively, indicating its good generalization ability. The proposed model achieved good performance, validated the predictive value of interim PET/CT, and holds promise for directing individualized clinical treatment. • The proposed multimodal models achieved accurate prediction of primary treatment failure in DLBCL patients. • Using an appropriate feature-level fusion strategy can make the same class close to each other regardless of the modal heterogeneity of the data source domain and positively impact the prediction performance. • Deep learning validated the predictive value of interim PET/CT in a way that exceeded human capabilities.

A novel method for steel bar all-stage pitting corrosion monitoring using the feature-level fusion of ultrasonic direct waves and coda waves

Corrosion monitoring of steel bars has drawn extensive attention in recent decades. Conventional ultrasonic method, utilizing direct waves to detect damage, is adequate for severe pitting corrosion but suffers from low sensitivity to incipient pitting corrosion. Coda wave technique, a very sensitive method to subtle changes in medium using later arrival wave packets, is innovatively introduced to monitor pitting corrosion of steel bars, especially in the early stages. The decorrelation coefficient (DC) values are calculated to quantify the variations of both direct waves and coda waves. To overcome the limitations of coda waves for severe pitting corrosion and remedy the low sensitivity of direct waves for incipient pitting corrosion, a feature-level data fusion strategy is proposed to integrate the two probing waves to monitor all-stage pitting corrosion of steel bars. The combination of direct waves and coda waves could exploit the complementary merits in various pitting corrosion configurations. The proposed feature-level fusion strategy of ultrasonic coda waves and direct waves intercepted from the same recorded signals opens a new perspective in all-stage pitting corrosion monitoring of steel bars and contributes a novel scheme for whole-process damage evaluation of structures.

Feature and decision-level fusion for schizophrenia detection based on resting-state fMRI data.

Mental disorders, especially schizophrenia, still pose a great challenge for diagnosis in early stages. Recently, computer-aided diagnosis techniques based on resting-state functional magnetic resonance imaging (Rs-fMRI) have been developed to tackle this challenge. In this work, we investigate different decision-level and feature-level fusion schemes for discriminating between schizophrenic and normal subjects. Four types of fMRI features are investigated, namely the regional homogeneity, voxel-mirrored homotopic connectivity, fractional amplitude of low-frequency fluctuations and amplitude of low-frequency fluctuations. Data denoising and preprocessing were first applied, followed by the feature extraction module. Four different feature selection algorithms were applied, and the best discriminative features were selected using the algorithm of feature selection via concave minimization (FSV). Support vector machine classifiers were trained and tested on the COBRE dataset formed of 70 schizophrenic subjects and 70 healthy subjects. The decision-level fusion method outperformed the single-feature-type approaches and achieved a 97.85% accuracy, a 98.33% sensitivity, a 96.83% specificity. Moreover, feature-fusion scheme resulted in a 98.57% accuracy, a 99.71% sensitivity, a 97.66% specificity, and an area under the ROC curve of 0.9984. In general, decision-level and feature-level fusion schemes boosted the performance of schizophrenia detectors based on fMRI features.