Search

Abstract Magnetic resonance imaging (MRI) is a non‐interposition imaging technique that provides rich anatomical and physiological information. Yet it is limited by the long imaging time. Recently, deep neural networks have shown potential to significantly accelerate MRI. However, most of these approaches ignore the correlation between adjacent slices in MRI image sequences. In addition, the existing deep learning‐based methods for MRI are mainly based on convolutional neural networks (CNNs). They fail to capture long‐distance dependencies due to the small receptive field. Inspired by the feature similarity in adjacent slices and impressive performance of Transformer for exploiting the long‐distance dependencies, a novel multislice texture transformer network is presented for undersampled MRI reconstruction (TTMRI). Specifically, the proposed TTMRI is consisted of four modules, namely the texture extraction, correlation calculation, texture transfer and texture synthesis. It takes three adjacent slices as inputs, in which the middle one is the target image to be reconstructed, and the other two are auxiliary images. The multiscale features are extracted by the texture extraction module and their inter‐dependencies are calculated by the correlation calculation module, respectively. Then the relevant features are transferred by the texture transfer module and fused by the texture synthesis module. By considering inter‐slice correlations and leveraging the Transformer architecture, the joint feature learning across target and adjacent slices are encouraged. Moreover, TTMRI can be stacked with multiple layers to recover more texture information at different levels. Extensive experiments demonstrate that the proposed TTMRI outperforms other state‐of‐the‐art methods in both quantitative and qualitative evaluationsions.

Clustering plays a major role in various disciplines including biological visual image classification. Various clustering algorithms have been introduced and developed. Among them, the density-based clustering algorithm achieves the best performance. The cellular neural network (CNN) based clustering algorithm (CNNCA) is a density-based method. It performs the clustering at a very high speed because the CNN can be implemented on a silicon chip. Nevertheless, the performance of the CNNCA is significantly degraded when facing clusters of complex structures, especially with significant density differences between clusters. In this paper, the local rules of the CNNs are modified. Moreover, the multi-spatial resolutions are employed. In particular, the values of minimum spatial resolution and the other parameters are adaptively designed. Hence, the iterative procedures can be applied to the remaining unassigned points. Furthermore, the conditional minimum distance rule is employed to assign the dispersed elements to the existing clusters. Different datasets have been utilized to evaluate the performance of our proposed method. Compared to the state-of-the-art clustering algorithms, the computer numerical simulation results show that our proposed method outperforms the existing state-of-the-art methods in higher accuracy and more robustness.

Abstract The traditional blood glucose estimation method requires to take the invasive measurements several times a day. Therefore, it has a high infection risk and the users are suffered from the pain. Moreover, the long term consumable cost is high. Recently, the wearable and non‐invasive blood glucose estimation approach has been proposed. However, due to the unreliability of the acquisition device, the presence of the noise and the variations of the acquisition environments, the obtained features and the reference blood glucose values are highly unreliable. Moreover, different subjects have different responses of the infrared light to the blood glucose. To address this issue, a polynomial fitting approach to smooth the obtained features or the reference blood glucose values has been proposed. In particular, the design of the coefficients in the polynomial is formulated as the various optimisation problems. First, the blood glucose values are estimated based on the individual optimisation approaches. Second, the absolute difference values between the estimated blood glucose values and the actual blood glucose values based on each optimisation approach are computed. Third, these absolute difference values for each optimisation approach are sorted in the ascending order. Fourth, for each sorted blood glucose value, the optimisation method corresponding to the minimum absolute difference value is selected. Fifth, the accumulate probability of each selected optimisation method is computed. If the accumulate probability of any selected optimisation method at a point is greater than a threshold value, then the accumulate probabilities of these three selected optimisation methods at that point are reset to zero. A range of the sorted blood glucose values are defined as that with the corresponding boundaries points being the previous reset point and this reset point. Hence, after performing the above procedures for all the sorted reference blood glucose values in the validation set, the regions of the sorted reference blood glucose values and the corresponding optimisation methods in these regions are determined. It is worth noting that the conventional lowpass denoising method was performed in the signal domain (either in the time domain or in the frequency domain), while the authors’ proposed method is performed in the feature space or the reference blood glucose space. Hence, the authors’ proposed method can further improve the reliability of the obtained feature values or the reference blood glucose values so as to improve the accuracy of the blood glucose estimation. Moreover, the individual modelling regression method has been employed here to suppress the effects of different users having different responses of the infrared light to the blood glucose values. The computer numerical simulation results show that the authors’ proposed method yields the mean absolute relative deviation (MARD) at 0.0930 and the percentage of the test data falling in the zone A of the Clarke error grid at 94.1176%.

The integration of Synthetic Aperture Radar (SAR) imaging technology with deep neural networks has experienced significant advancements in recent years. Yet, the scarcity of high-quality samples and the difficulty of extracting prior information from SAR data have experienced limited progress in this domain. This study introduces an innovative sparse SAR imaging approach using a self-supervised non-local asymmetric pixel-shuffle blind spot network. This strategy enables the network to be trained without labeled samples, thus solving the problem of the scarcity of high-quality samples. Through asymmetric pixel-shuffle downsampling (AP) operation, the spatial correlation between pixels is broken so that the blind spot network can adapt to the actual scene. The network also incorporates a non-local module (NLM) into its blind spot architecture, enhancing its capability to analyze a broader range of information and extract more comprehensive prior knowledge from SAR data. Subsequently, Plug and Play (PnP) technology is used to integrate the trained network into the sparse SAR imaging model to solve the regularization term problem. The optimization of the inverse problem is achieved through the Alternating Direction Method of Multipliers (ADMM) algorithm. The experimental results of the unlabeled samples demonstrate that our method significantly outperforms traditional techniques in reconstructing images across various regions.

In the Internet of Things (IoT) communications, visual data are frequently processed among intelligent devices using artificial intelligence algorithms, replacing humans for analysis and decision-making while only occasionally requiring human scrutiny. However, due to high redundancy of compressive encoders, existing image coding solutions for machine vision are inefficient at runtime. To balance the rate-accuracy performance and efficiency of image compression for machine vision while attaining high-quality reconstructed images for human vision, this paper introduces a novel slimmable multi-task compression framework for human and machine vision in visual IoT applications. Firstly, image compression for human and machine vision under the constraint of bandwidth, latency, and computational resources is modeled as a multi-task optimization problem. Secondly, slimmable encoders are employed for multiple human and machine vision tasks in which the parameters of the sub-encoder for machine vision tasks are shared among all tasks and jointly learned. Thirdly, to solve the feature match between latent representation and intermediate features of deep vision networks, feature transformation networks are introduced as decoders of machine vision feature compression. Finally, the proposed framework is successfully applied to human and machine vision tasks’ scenarios, e.g., object detection and image reconstruction. Experimental results show that the proposed method outperforms baselines and other image compression approaches on machine vision tasks with higher efficiency (shorter latency) in two vision tasks’ scenarios while retaining comparable quality on image reconstruction.

Multivariate ridge regression (MR), linear discriminant analysis (LDA) and extreme learning machine (ELM) have been widely used in hyperspectral image (HSI) classification. However, these methods do not consider the influence of noise in HSIs, spatial information, and the internal relationship between samples. As a result, the sample distribution is not ideal and the classification effect cannot be improved. This article extends LDA and MR to the field of tensors, that can not only use the spatial information of the sample, but also can make the distribution of homogeneous samples more concentrated. Besides, this article analyzes the relationship between the number of neurons in the hidden layer of ELM and the classification accuracy. Finally, singular spectral analysis (SSA) is chosen to improve classification accuracy. The tensor discriminant ridge regression model with ELM and SSA for HSI classification is proposed. Experiments show compared with tensor-based classifiers, ELM and other state-of-the-art methods, the proposed method is efficient and effective.

Abstract Consider general minimum variance distortionless response (MVDR) robust adaptive beamforming problems based on the optimal estimation for both the desired signal steering vector and the interference‐plus‐noise covariance (INC) matrix. The optimal robust adaptive beamformer design problem is an array output power maximization problem, subject to three constraints on the steering vector, namely, a (convex or nonconvex) quadratic constraint ensuring that the direction‐of‐arrival (DOA) of the desired signal is separated from the DOA region of all linear combinations of the interference steering vectors, a double‐sided norm constraint, and a similarity constraint; as well as a ball constraint on the INC matrix, which is centered at a given data sample covariance matrix. To tackle the nonconvex problem, a new tightened semidefinite relaxation (SDR) approach is proposed to output a globally optimal solution; otherwise, a sequential convex approximation (SCA) method is established to return a locally optimal solution. The simulation results show that the MVDR robust adaptive beamformers based on the optimal estimation for the steering vector and the INC matrix have better performance (in terms of, e.g., the array output signal‐to‐interference‐plus‐noise ratio) than the existing MVDR robust adaptive beamformers by the steering vector estimation only.

Oceanic targets, including ripples, islands, vessels, and coastlines, display distinct sparse characteristics, rendering the ocean a significant arena for sparse Synthetic Aperture Radar (SAR) imaging rooted in sparse signal processing. Deep neural networks (DNNs), a current research emphasis, have, when integrated with sparse SAR, attracted notable attention for their exceptional imaging capabilities and high computational efficiency. Yet, the efficiency of traditional unfolding techniques is impeded by their architecturally inefficient design, which curtails their information transmission capacity and consequently detracts from the quality of reconstruction. This paper unveils a novel Memory-Augmented Deep Unfolding Network (MADUN) for SAR imaging in marine environments. Our methodology harnesses the synergies between deep learning and algorithmic unfolding, enhanced with a memory component, to elevate SAR imaging’s computational precision. At the heart of our investigation is the incorporation of High-Throughput Short-Term Memory (HSM) and Cross-Stage Long-Term Memory (CLM) within the MADUN framework, ensuring robust information flow across unfolding stages and solidifying the foundation for deep, long-term informational correlations. Our experimental results demonstrate that our strategy significantly surpasses existing methods in enhancing the reconstruction of sparse marine scenes.

Abstract In this paper, the application of a hybrid representation in wide‐angle synthetic aperture radar (WASAR) imaging is investigated, addressing the challenge of achieving sparse representation in the presence of complex electromagnetic scattering characteristics and highly anisotropic targets. A convolutional neural network is utilized to represent the two‐dimensional data within each azimuth aperture, while employing blind compressed sensing (BCS) to achieve sparse representation across different azimuth apertures. Convolutional neural network (CNN) excels in learning spatial hierarchies and local dependencies of two‐dimensional data but requires a large amount of training data. Isotropic targets within each azimuth aperture can be used for training conventional SAR data, while acquiring training samples for anisotropic targets poses challenges. To address this issue, BCS are integrated into WASAR imaging, eliminating the need for additional training data. By integrating these methods, a novel approach called hybrid‐WASAR is proposed, which incorporates two regularization terms into WASAR imaging and employs the alternating direction method of multipliers to iteratively solve the imaging model. Compared to traditional WASAR imaging techniques, hybrid‐WASAR improves the accuracy of reconstructing the backscattering coefficients of the targets, leading to a significant enhancement in overall imaging quality.

The integration of deep neural networks into sparse synthetic aperture radar (SAR) imaging is explored to enhance SAR imaging performance and reduce the system’s sampling rate. However, the scarcity of training samples and mismatches between the training data and the SAR system pose significant challenges to the method’s further development. In this paper, we propose a novel SAR imaging approach based on deep image prior powered by RED (DeepRED), enabling unsupervised SAR imaging without the need for additional training data. Initially, DeepRED is introduced as the regularization technique within the sparse SAR imaging model. Subsequently, variable splitting and the alternating direction method of multipliers (ADMM) are employed to solve the imaging model, alternately updating the magnitude and phase of the SAR image. Additionally, the SAR echo simulation operator is utilized as an observation model to enhance computational efficiency. Through simulations and real data experiments, we demonstrate that our method maintains imaging quality and system downsampling rate on par with deep-neural-network-based sparse SAR imaging but without the requirement for training data.

Abstract Prehypertension is a new risky disease defined in the seventh report issued by the Joint National Commission. Hence, detecting prehypertension in time plays a very important role in protecting human lives. This study proposes a method for categorising blood pressure values into two classes, namely the class of healthy blood pressure values and the class of prehypertension blood pressure values, as well as estimating the blood pressure values continuously only by employing photoplethysmograms. First, the denoising of photoplethysmograms is performed via a discrete cosine transform approach. Then, the features of the photoplethysmograms in both the time domain and the frequency domain are extracted. Next, the feature vectors are categorised into the two classes of blood pressure values by a multi‐model fusion of the classifiers. Here, the support vector machine, the random forest and the K‐nearest neighbour classifier are employed for performing the fusion. There are two types of blood pressure values. They are the systolic blood pressure values and the diastolic blood pressure values. For each class and each type of blood pressure values, support vector regression is used to estimate the blood pressure values. Since different classes and different types of blood pressure values are considered separately, the proposed method achieves an accurate estimation. The computed numerical simulation results show that the proposed method based on the multi‐model fusion of the classifiers achieves both higher classification accuracy and higher regression accuracy than the individual classification methods.

With the advent of autonomous vehicle applications, the importance of LiDAR point cloud 3D object detection cannot be overstated. Recent studies have demonstrated that methods for aggregating features from voxels can accurately and efficiently detect objects in large, complex 3D detection scenes. Nevertheless, most of these methods do not filter background points well and have inferior detection performance for small objects. To ameliorate this issue, this paper proposes an Attention-based and Multiscale Feature Fusion Network (AMFF-Net), which utilizes a Dual-Attention Voxel Feature Extractor (DA-VFE) and a Multi-scale Feature Fusion (MFF) Module to improve the precision and efficiency of 3D object detection. The DA-VFE considers pointwise and channelwise attention and integrates them into the Voxel Feature Extractor (VFE) to enhance key point cloud information in voxels and refine more-representative voxel features. The MFF Module consists of self-calibrated convolutions, a residual structure, and a coordinate attention mechanism, which acts as a 2D Backbone to expand the receptive domain and capture more contextual information, thus better capturing small object locations, enhancing the feature-extraction capability of the network and reducing the computational overhead. We performed evaluations of the proposed model on the nuScenes dataset with a large number of driving scenarios. The experimental results showed that the AMFF-Net achieved 62.8% in the mAP, which significantly boosted the performance of small object detection compared to the baseline network and significantly reduced the computational overhead, while the inference speed remained essentially the same. AMFF-Net also achieved advanced performance on the KITTI dataset.

It is worth noting that a graphical equalizer is a linear time-invariant system. Hence, its input-output relationship can be modeled by a transfer function. The graphical equalizer is realized by a parallel connection of a set of filters called a filter bank. Each filter in the filter bank is designed to equalize a certain frequency band of the audio signal. There are two types of graphic equalizers. The first type of graphical equalizers is based on a finite impulse response (FIR) filter bank. However, it requires very high order FIR filters to achieve a certain degree of equalization precision. As a result, the required computational power is very high. Also, the output of the graphical equalizer is suffered from a very long time delay. The second type of graphical equalizers is based on an infinite impulse response (IIR) filter bank. It can address the above issues. That is, its implementation cost is low. To design each IIR filter, the total absolute error between the ideal equalization and the actual equalization in the passband of the IIR filter is minimized subject to the constraints imposed on the stability criterion of the IIR filter as well as on the maximum modulus error between the ideal equalization and the actual equalization in the corresponding passband. This design problem is a functional inequality constrained optimization problem which consists of an infinite number of inequality constraints. By performing the downsampling in the frequency domain, the infinite constrained optimization problem becomes the finite constrained optimization problem. By using the genetic algorithm to find the solution of this optimization problem, the coefficients of all the IIR filters are obtained. The computer numerical simulation results show that the equalizer designed by our proposed method yields a good equalization performance.

Attention refers to the human psychological ability to focus on doing an activity. The attention assessment plays an important role in diagnosing attention deficit hyperactivity disorder (ADHD). In this paper, the attention assessment is performed via a classification approach. First, the single-channel electroencephalograms (EEGs) are acquired from various participants when they perform various activities. Then, fast Fourier transform (FFT) is applied to the acquired EEGs, and the high-frequency components are discarded for performing denoising. Next, empirical mode decomposition (EMD) is applied to remove the underlying trend of the signals. In order to extract more features, singular spectrum analysis (SSA) is employed to increase the total number of the components. Finally, some typical models such as the random forest-based classifier, the support vector machine (SVM)-based classifier, and the back-propagation (BP) neural network-based classifier are used for performing the classifications. Here, the percentages of the classification accuracies are employed as the attention scores. The computer numerical simulation results show that our proposed method yields a higher classification performance compared to the traditional methods without performing the EMD and SSA.

Because conventional PCANET approach is that the conventional PCANET performs the PCA for all the segments of all the training pixel vectors, and this does not capture the difference between different segments of the same training pixel vectors, classification accuracy is not high. This paper proposes to employ a regional principal component analysis network with the rolling guidance filter (RPCANET_RGF) for performing the hyperspectral image (HSI) classification with few training samples. Regional principal component analysis network (RPCANET) proposed in this paper performs the PCA for each segment of all training pixel vectors. Besides, the rolling guidance filter (RGF) is used to remove the spatial noise and to enhance the edges of the HSIs. Different from the conventional convolutional neural networks (CNNs), the coefficients of the filters are obtained by performing the principal component analysis (PCA) on the regional segments of HSIs. This approach is also different from the conventional principal component analysis network (PCANET). Here, different segments of the same pixel image are processed by different Filters. Since the RPCANET_RGF is a general learning method that obtains the filter coefficients directly from the HSIs, the back propagation based training is not required. Hence, the RPCANET_RGF requires a less computational power for performing the training compared to the CNN. Besides, as the RPCANET_RGF can make use of both the spectral information and the spatial information for performing the classification, the computer numerical simulation results show that the classification accuracy achieved by the RPCANET_RGF is higher than that by the conventional PCANET and other state of the art methods.

Existing works on leather surface defects are mainly focused on the defect detection without performing the classification automatically. On the other hand, the manual classification has the disadvantages including a misjudgment, the occurrence of cumbersome labors and a high cost. To address these issues, this paper proposes a framework for classifying the leather surface defects based on a parameter optimized residual network. In this framework, an ultra high definition imaging system is first utilized to obtain the images of the leather surface. Then, two important network parameters are optimized. They are the size of the image data set and the size of the sliding patch window. Here, the size of the image data set is determined by achieving the tradeoffs between the evaluated workload and the classification accuracy. On the other hand, the size of the sliding patch window is obtained by the least squares method. In order to mitigate the limitation of the gradient disappearance in a deep network, a residual module is introduced to the proposed framework. Finally, the defect classification is performed based on the extracted features using multi-layer convolution and pooling operations. Computer numerical simulation results show that the leather surface defects can be classified effectively by the proposed framework. Also, the classification accuracy can reach 94.6%.

Automatic segmentation of the cancerous esophagus in computed tomography (CT) images is a computer-assisted method that can improve the efficiency of the diagnosis and treatment. Due to the diversity of the cancer stage and location, the anatomical structure of the cancerous esophagus is various. Moreover, the low contrast against surrounding tissues leads to a blurry boundary of the cancerous esophagus. Therefore, existing segmentation networks cannot achieve satisfactory results in automatic segmentation of the cancerous esophagus. In this article, we propose a novel 2.5D segmentation network named Eso-Net for the cancerous esophagus based on an encoder-decoder architecture. A 3D enhancement filter called Multi-Structure Response Filter (MSRF) is designed to extract 3D structural information as prior knowledge. Furthermore, dilated convolutions and residual connections are employed in the convolutional blocks of Eso-Net for multi-scale feature learning. With 3D structural priors, Prior Attention Modules (PAM) are incorporated into the network to facilitate the transmission of relevant spatial information. The experiments are conducted on the dataset from 30 esophageal cancer patients, and we report an 84.839% dice similarity coefficient, an 85.955% precision, an 83.752% sensitivity, and a 2.583mm Hausdorff distance. The experimental results demonstrate that the proposed method outperforms other existing segmentation networks in this task and can effectively assist doctors in the diagnosis and treatment of esophageal cancer.

The conventional singular spectrum analysis is to divide a signal into segments where there is only one non-overlapping point between two consecutive segments. By putting these segments into the columns of a matrix and performing the singular value decomposition on the matrix, various one dimensional singular spectrum analysis vectors are obtained. Since different one dimensional singular spectrum analysis vectors represent different parts of the signal such as the trend part, the oscillation part and the noise part of the signal, the singular spectrum analysis plays a very important role in the nonlinear and adaptive signal analysis. However, as the length of each one dimensional singular spectrum analysis vector is the same as that of the original signal, there is a redundancy among these one dimensional singular spectrum analysis vectors. In order to reduce the required computational power and the required units for the memory storage for performing the singular spectrum analysis, this article proposes a method to reduce the total number of the elements of all the one dimensional singular spectrum analysis vectors. In particular, the length of the shift block for generating the trajectory matrix is increased from one to a positive integer greater than one under a certain criterion. In this case, the total number of the columns of the trajectory matrix is reduced. As a result, the total number of the off-diagonals of all the two dimensional singular spectrum analysis matrices is reduced. Hence, the total number of the elements of all the one dimensional singular spectrum analysis vectors is reduced. In order to guarantee exact perfect reconstruction, this article reformulates the de-Hankelization process. In particular, the first element of the off-diagonals of all the two dimensional singular spectrum analysis matrices is taken as the elements of the one dimensional singular spectrum analysis vectors. Exact perfect reconstruction condition is derived. Simulations show that exact perfect reconstruction can be achieved while the total number of the elements of all the one dimensional singular spectrum analysis vectors is significantly reduced.

Recently, deep learning-based image quality enhancement models have been proposed to improve the perceptual quality of distorted synthesized views impaired by compression and the Depth Image-Based Rendering (DIBR) process in a multi-view video system. However, due to the lack of Multi-view Video plus Depth (MVD) data, the training data for quality enhancement models is small, which limits the performance and progress of these models. Augmenting the training data to enhance the synthesized view quality enhancement (SVQE) models is a feasible solution. In this paper, a deep learning-based SVQE model using more synthetic synthesized view images (SVIs) is suggested. To simulate the irregular geometric displacement of DIBR distortion, a random irregular polygon-based SVI synthesis method is proposed based on existing massive RGB/RGBD data, and a synthetic synthesized view database is constructed, which includes synthetic SVIs and the DIBR distortion mask. Moreover, to further guide the SVQE models to focus more precisely on DIBR distortion, a DIBR distortion mask prediction network which could predict the position and variance of DIBR distortion is embedded into the SVQE models. The experimental results on public MVD sequences demonstrate that the PSNR performance of the existing SVQE models, e.g., DnCNN, NAFNet, and TSAN, pre-trained on NYU-based synthetic SVIs could be greatly promoted by 0.51-, 0.36-, and 0.26 dB on average, respectively, while the MPPSNRr performance could also be elevated by 0.86, 0.25, and 0.24 on average, respectively. In addition, by introducing the DIBR distortion mask prediction network, the SVI quality obtained by the DnCNN and NAFNet pre-trained on NYU-based synthetic SVIs could be further enhanced by 0.02- and 0.03 dB on average in terms of the PSNR and 0.004 and 0.121 on average in terms of the MPPSNRr.

In order to address the problem of achieving the poor performance of a single-channel receiver operated at a low signal-to-noise ratio (SNR) based on the existing source number estimation method, an effective source enumeration approach for the single-channel receiver operated at a low SNR is proposed in this paper. The proposed method is based on the empirical mode decomposition (EMD) with the auto-correlation coefficient matrix (ACCM) and the jackknifing method. First, the received single-channel signal is decomposed into several intrinsic mode functions and a residue via the EMD. Both the components and the original signal are employed to simulate the signals obtained from a pseudo-uniform linear array (ULA). After applying the multiple jackknifing methods to the signals obtained from the pseudo-ULA without complete overlap, we acquire a series of subsample datasets. Consequently, the ACCM instead of the conventional auto-covariance matrix is constructed for each subsample. After performing the eigenvalue decomposition of all the ACCMs, we acquire several sets of eigenvalues. For each set of eigenvalues, either moving increment criterion or moving standard deviation criterion is employed to determine the source number. Since more than one ACCM is employed because of using the multiple jackknifing methods, we have the statistics for the estimation of each ACCM. Then, we take the one that occurs the most frequently as the final result of source enumeration. The experiments validate the proposed method and show the excellent performance of our proposed algorithm. Finally, we also present the optimal scheme for selecting the important parameters, such as the jackknifing ratio and the jackknifing times in the approach.

This paper proposes a two dimensional quaternion valued singular spectrum analysis based method for enhancing the hyperspectral image. Here, the enhancement is for performing the object recognition, but neither for improving the visual quality nor suppressing the artifacts. In particular, the two dimensional quaternion valued singular spectrum analysis components are selected in such a way that the ratio of the interclass separation to the intraclass separation of the pixel vectors is maximized. Next, the support vector machine is employed for performing the object recognition. Compared to the conventional two dimensional real valued singular spectrum analysis based method where only the pixels in a color plane is exploited, the two dimensional quaternion valued singular spectrum analysis based method fuses four color planes together for performing the enhancement. Hence, both the spatial information among the pixels in the same color plane and the spectral information among various color planes are exploited. The computer numerical simulation results show that the overall classification accuracy based on our proposed method is higher than the two dimensional real valued singular spectrum analysis based method, the three dimensional singular spectrum analysis based method, the multivariate two dimensional singular spectrum analysis based method, the median filtering based method, the principal component analysis based method, the Tucker decomposition based method and the hybrid spectral convolutional neural network (hybrid SN) based method.

This paper aims to develop an activity recognition algorithm to allow parents to monitor their children at home after school. A common method used to analyze electroencephalograms is to use infinite impulse response filters to decompose the electroencephalograms into various brain wave components. However, nonlinear phase distortions will be introduced by these filters. To address this issue, this paper applies empirical mode decomposition to decompose the electroencephalograms into various intrinsic mode functions and categorize them into four groups. In addition, common features used to analyze electroencephalograms are energy and entropy. However, because there are only two features, the available information is limited. To address this issue, this paper extracts 11 different physical quantities from each group of intrinsic mode functions, and these are employed as the features. Finally, this paper uses the random forest to perform activity recognition. It is worth noting that the conventional approach for performing activity recognition is based on a single type of signal, which limits the recognition performance. In this paper, a multi-modal system based on electroencephalograms, image sequences, and motion signals is used for activity recognition. The numerical simulation results show that the percentage accuracies based on three types of signal are higher than those based on two types of signal or the individual signals. This demonstrates the advantages of using the multi-modal approach for activity recognition. In addition, our proposed empirical mode decomposition-based method outperforms the conventional filtering-based method. This demonstrates the advantages of using the nonlinear and adaptive time frequency approach for activity recognition.

This paper is an extension of the existing works on the frequency-domain-based bit flipping control strategy for stabilizing the single-bit high-order interpolative sigma delta modulator. In particular, this paper proposes the implementation and performs the performance evaluation of the control strategy. For the implementation, a frequency detector is used to detect the resonance frequencies of the input sequence of the sigma delta modulator. Then, a neural-network-based controller is used for finding the solution of the integer programming problem. Finally, the buffers and the combinational logic gates as well as an inverter are used for implementing the proposed control strategy. For the performance evaluation, the stability region in terms of the input dynamical range is evaluated. It is found that the control strategy can significantly increase the input dynamical range from 0.24 to 0.58. Besides, the control strategy can be applied to a wider class of the input signals compared to the clipping method.

This paper proposes a framework combining the complementary ensemble empirical mode decomposition with both the independent component analysis and the non-negative matrix factorization for estimating both the heart rate and the respiratory rate from the photoplethysmography (PPG) signal. After performing the complementary ensemble empirical mode decomposition on the PPG signal, a finite number of intrinsic mode functions are obtained. Then, these intrinsic mode functions are divided into two groups to perform the further analysis via both the independent component analysis and the non-negative matrix factorization. The surrogate cardiac signal related to the heart activity and another surrogate respiratory signal related to the respiratory activity are reconstructed to estimate the heart rate and the respiratory rate, respectively. Finally, different records of signals acquired from the Medical Information Mart for Intensive Care database downloaded from the Physionet Automated Teller Machine (ATM) data bank are employed for demonstrating the outperformance of our proposed method. The results show that our proposed method outperforms both the digital filtering approach and the conventional empirical mode decomposition based methods in terms of reconstructing both the surrogate cardiac signal and the respiratory signal from the PPG signal as well as both achieving the higher accuracy and the higher reliability for estimating both the heart rate and the respiratory rate.

The novelty and the contribution of this paper consists of applying an iterative joint singular spectrum analysis and low-rank decomposition approach for suppressing the spikes in an electroencephalogram. First, an electroencephalogram is filtered by an ideal lowpass filter via removing its discrete Fourier transform coefficients outside the δ wave band, the θ wave band, the α wave band, the β wave band and the γ wave band. Second, the singular spectrum analysis is performed on the filtered electroencephalogram to obtain the singular spectrum analysis components. The singular spectrum analysis components are sorted according to the magnitudes of their corresponding eigenvalues. The singular spectrum analysis components are sequentially added together starting from the last singular spectrum analysis component. If the variance of the summed singular spectrum analysis component under the unit energy normalization is larger than a threshold value, then the summation is terminated. The summed singular spectrum analysis component forms the first scale of the electroencephalogram. The rest singular spectrum analysis components are also summed up together separately to form the residue of the electroencephalogram. Next, the low-rank decomposition is performed on the residue of the electroencephalogram to obtain both the low-rank component and the sparse component. The low-rank component is added to the previous scale of the electroencephalogram to obtain the next scale of the electroencephalogram. Finally, the above procedures are repeated on the sparse component until the variance of the current scale of the electroencephalogram under the unit energy normalization is larger than another threshold value. The computer numerical simulation results show that the spike suppression performance based on our proposed method outperforms that based on the state-of-the-art methods.

In any diabetic retinopathy screening program, about two-thirds of patients have no retinopathy. However, on average, it takes a human expert about one and a half times longer to decide an image is normal than to recognize an abnormal case with obvious features. In this work, we present an automated system for filtering out normal cases to facilitate a more effective use of grading time. The key aim with any such tool is to achieve high sensitivity and specificity to ensure patients' safety and service efficiency. There are many challenges to overcome, given the variation of images and characteristics to identify. The system combines computed evidence obtained from various processing stages, including segmentation of candidate regions, classification and contextual analysis through Hidden Markov Models. Furthermore, evolutionary algorithms are employed to optimize the Hidden Markov Models, feature selection and heterogeneous ensemble classifiers. In order to evaluate its capability of identifying normal images across diverse populations, a population-oriented study was undertaken comparing the software's output to grading by humans. In addition, population based studies collect large numbers of images on subjects expected to have no abnormality. These studies expect timely and cost-effective grading. Altogether 9954 previously unseen images taken from various populations were tested. All test images were masked so the automated system had not been exposed to them before. This system was trained using image subregions taken from about 400 sample images. Sensitivities of 92.2% and specificities of 90.4% were achieved varying between populations and population clusters. Of all images the automated system decided to be normal, 98.2% were true normal when compared to the manual grading results. These results demonstrate scalability and strong potential of such an integrated computational intelligence system as an effective tool to assist a grading service.