Benchmark Databases Research Articles

Semantic segmentation of retinal exudates using a residual encoder-decoder architecture in diabetic retinopathy.

Exudates are a common sign of diabetic retinopathy, which is a disease that affects the blood vessels in the retina. Early detection of exudates is critical to avoiding vision problems through continuous screening and treatment. In traditional clinical practice, the involved lesions are manually detected using photographs of the fundus. However, this task is cumbersome and time-consuming and requires intense effort due to the small size of the lesion and the low contrast of the images. Thus, computer-assisted diagnosis of retinal disease based on the detection of red lesions has been actively explored recently. In this paper, we present a comparison of deep convolutional neural network (CNN) architectures and propose a residual CNN with residual skip connections to reduce the parameter for the semantic segmentation of exudates in retinal images. A suitable image augmentation technique is used to improve the performance of network architecture. The proposed network can robustly segment exudates with high accuracy, which makes it suitable for diabetic retinopathy screening. A comparative performance analysis of three benchmark databases: E-ophtha, DIARETDB1, and Hamilton Ophthalmology Institute's Macular Edema, is presented. The proposed method achieves a precision of 0.95, 0.92, 0.97, accuracy of 0.98, 0.98, 0.98, sensitivity of 0.97, 0.95, 0.95, specificity of 0.99, 0.99, 0.99, and area under the curve of 0.97, 0.94, and 0.96, respectively. RESEARCH HIGHLIGHTS: The research focuses on the detection and segmentation of exudates in diabetic retinopathy, a disease affecting the retina. Early detection of exudates is important to avoid vision problems and requires continuous screening and treatment. Currently, manual detection is time-consuming and requires intense effort. The authors compare qualitative results of the state-of-the-art convolutional neural network (CNN) architectures and propose a computer-assisted diagnosis approach based on deep learning, using a residual CNN with residual skip connections to reduce parameters. The proposed method is evaluated on three benchmark databases and demonstrates high accuracy and suitability for diabetic retinopathy screening.

Truncated attention-aware proposal networks with multi-scale dilation for temporal action detection

Detecting actions temporally in untrimmed videos is very challenging, and it accomplishes action classification and localization simultaneously. Capturing the relations among action proposals (i.e., candidate video segments) is of vital importance. While there have been several attempts to encode such relations, they neglect the adverse effects of those irrelevant or negative relations among proposals. Besides, there is a crucial fact that action durations are flexible in videos, which has not been well explored. For the former, we develop a truncated attention mechanism that learns positive proposal relations by dynamically adjusting edge weights of proposal nodes in a graph, and construct the proposal network model using graph convolution networks to suppress disadvantageous relations of proposal pairs by truncating negative attention scores. For the latter, we devise a light multi-scale dilation module shared by all proposals to handle different action durations by enlarging temporal receptive field, thus capturing temporal context to increase the representation capacity of proposals. Unifying these considerations, we present the Multi-scale Dilation based Truncated Attention Proposal Network (MD-TAPN) model for temporal action detection. Our model achieves state-of-the-art performances of detecting actions on two benchmark databases, and especially it outperforms the most competitive method by a significant gain of 3.6% mAP at tIoU0.5 on THUMOS14.

Multi-source domain adaptation with spatio-temporal feature extractor for EEG emotion recognition

Electroencephalogram (EEG) emotion recognition suffers from cross-domain identification difficulties mainly caused by cross-subject and cross-session differences. Inspired by domain adaptation, the study constructs a novel multi-source domain adaptation with a Spatio-temporal feature extractor (MSDA-SFE) for EEG emotion recognition, which could reduce the signal differences between subjects and between sessions to make the target domain features align with the source domain features easily. In common feature learning, a Spatiotemporal feature extraction module is constructed to acquire domain-invariant characteristics of the source and target domains in which one subject is selected randomly as the target domain, and the remaining subjects are source domains. Whereafter, the obtained target domain features are paired with other source domain features respectively to generate N-1 pairs of concatenated features. And then these features are translated into N-1 domain-specific EEG features through N-1 parallel branches, where the target domain features are better aligned in the latent space, and the classification boundary of target samples is obtained by minimizing the disparity between branches. Finally, domain-specific features are fed to respective classifiers to get N-1 predictions, and their mean is put as the final result. Extensive experiments are implemented on SEED, SEED-IV, and DEAP benchmark databases. And, results indicate that the MSDA-SFE network outperforms the other comparison methods. Besides, the additional experiment also illustrates that the method has a powerful generalization capacity.

Probabilistic quaternion collaborative representation and its application to robust color face identification

Quaternion representation (QR) has been shown to be an effective tool for image processing by modelling color images as quaternion matrices. However, previous QR based classification (QRC) methods overlook the labels of training data and may lead to sub-optimal results. In this article, we propose a Probabilistic Quaternion Collaborative Representation (ProQCR) method based on maximum likelihood estimation with application to color face identification. The proposed ProQCR method has the following merits: (1) Full supervision: In the representation and classification processes, the ProQCR method exploits the label information of training data, in contrast to prior methods using such supervision information only in the classification step. (2) Multi-channel integrality: ProQCR represents multiple color channels of each test color image as a quaternion linear combination of training data in a holistic manner. (3) Robustness: Equipped with the quaternion Huber loss function, ProQCR is robust against gross corruption in the test data. Experiments on benchmark databases demonstrate the effectiveness and robustness of ProQCR for color face identification.

Nuclear data for fusion: inventory validation successes and future needs

Nuclear data, describing neutron reaction probabilities (cross sections) and decay behaviour, are critical to the design and operation of fusion experiments and future fusion power plants. Equally vital, are the inventory codes that use the data to predict neutron-induced activation and transmutation of materials, which will define the radiological hazards that must be managed during reactor operation and decommissioning. Transmutation, including gas production, combined with the neutron-induced displacement damage, will also cause the properties of materials to degrade, for example through swelling and embrittlement, eventually limiting the lifetime of components. Thus validated and accurate nuclear data and inventory codes are essential. For data validation there are decay heat measurements performed at FNS in Japan more than 20 years ago. The experiments produced an invaluable database for benchmarking of nuclear data libraries; the latest versions of several international libraries perform well against this data during tests with the FISPACT-II inventory code, although there is still scope for improvement. A recent attempt to provide fusion-relevant validation based on γ-spectroscopy data from neutron-irradiated material samples tests produced predictions for short-lived (several hours or less) radionuclides. The detailed analysis performed for molybdenum demonstrates how these data could eventually provide a new benchmark, and also illustrates the potential benefits of further experiments targeting the longer-lived radionuclides relevant to maintenance and decommissioning timescales. There are also some successful tests of transmutation predictions with FISPACT-II. These direct validations of inventory simulations are critical for lifetime predictions and future experiments should learn lessons from the examples described for tungsten, which demonstrate the importance of an accurate description of the neutron spectrum in experiments. More novel experimental techniques are needed to measure helium production in materials such as Fe and C, but the need to validate the nuclear data evaluations used by simulations should motivate future experimental efforts.

Emotion Recognition Using a Novel Granger Causality Quantifier and Combined Electrodes of EEG.

Electroencephalogram (EEG) connectivity patterns can reflect neural correlates of emotion. However, the necessity of evaluating bulky data for multi-channel measurements increases the computational cost of the EEG network. To date, several approaches have been presented to pick the optimal cerebral channels, mainly depending on available data. Consequently, the risk of low data stability and reliability has increased by reducing the number of channels. Alternatively, this study suggests an electrode combination approach in which the brain is divided into six areas. After extracting EEG frequency bands, an innovative Granger causality-based measure was introduced to quantify brain connectivity patterns. The feature was subsequently subjected to a classification module to recognize valence-arousal dimensional emotions. A Database for Emotion Analysis Using Physiological Signals (DEAP) was used as a benchmark database to evaluate the scheme. The experimental results revealed a maximum accuracy of 89.55%. Additionally, EEG-based connectivity in the beta-frequency band was able to effectively classify dimensional emotions. In sum, combined EEG electrodes can efficiently replicate 32-channel EEG information.

Learning transferable non-negative feature representation for facial expression recognition

Conventional facial expression recognition (FER) methods often assume that the training and test procedures are carried out on a single database, without considering the cross-database problem due to the difference in ages, collecting devices, nationalities, etc. To tackle the above challenge, we propose a novel transferable non-negative feature representation (TNFR) method in this work. The core ideas of TNFR are two-fold. First, we utilize the basis matrix as a transformation matrix to reduce the distribution discrepancy between two databases. Second, we utilize the label information to improve the discrimination ability of the model. In addition, we replace the original label matrix with a block-label matrix to make the model more flexible. Experiments on four benchmark databases are provided to demonstrate the effectiveness of the proposed TNFR method in comparison with state-of-the-art methods.

AFSFusion: An Adjacent Feature Shuffle Combination Network for Infrared and Visible Image Fusion

To obtain fused images with excellent contrast, distinct target edges, and well-preserved details, we propose an adaptive image fusion network called the adjacent feature shuffle-fusion network (AFSFusion). The proposed network adopts a UNet-like architecture and incorporates key refinements to enhance network architecture and loss functions. Regarding the network architecture, the proposed two-branch adjacent feature fusion module, called AFSF, expands the number of channels to fuse the feature channels of several adjacent convolutional layers in the first half of the AFSFusion, enhancing its ability to extract, transmit, and modulate feature information. We replace the original rectified linear unit (ReLU) with leaky ReLU to alleviate the problem of gradient disappearance and add a channel shuffling operation at the end of AFSF to facilitate information interaction capability between features. Concerning loss functions, we propose an adaptive weight adjustment (AWA) strategy to assign weight values to the corresponding pixels of the infrared (IR) and visible images in the fused images, according to the VGG16 gradient feature response of the IR and visible images. This strategy efficiently handles different scene contents. After normalization, the weight values are used as weighting coefficients for the two sets of images. The weighting coefficients are applied to three loss items simultaneously: mean square error (MSE), structural similarity (SSIM), and total variation (TV), resulting in clearer objects and richer texture detail in the fused images. We conducted a series of experiments on several benchmark databases, and the results demonstrate the effectiveness of the proposed network architecture and the superiority of the proposed network compared to other state-of-the-art fusion methods. It ranks first in several objective metrics, showing the best performance and exhibiting sharper and richer edges of specific targets, which is more in line with human visual perception. The remarkable enhancement in performance is ascribed to the proposed AFSF module and AWA strategy, enabling balanced feature extraction, fusion, and modulation of image features throughout the process.

Turning waste into wealth: Person identification by emotion‐disturbed electrocardiogram

AbstractThe issue of electrocardiogram (ECG)‐based person identification has attracted intense research interests nowadays. Different than existing related researches that advocate accentuating useful information and attenuating noisy artefacts in sensor data processing, A novel strategy of ‘turning waste into wealth’ is proposed to exploit the new discriminative information from the relationship between noise disturbance and signal data for this issue. Specifically, the authors design a new and simple method, the Set‐Group Distance Measure, based on the suitable fusion of multiple minority‐based distance measurements, whose power has initially been discovered for the issue. This method takes advantage of the collaborative variation information from the relative relationship, which is named as ‘relative information’, between different types of emotional noise disturbances and ECG signal data, to tackle the problem of large intra‐class variation but small inter‐class difference during identification. Experimental results have demonstrated the reasonability, effectiveness, robustness, efficiency and practicability of the proposed method upon public benchmark databases. This proposal not only provides technological inspirations for the further study in ECG‐based person identification, but also shows a fresh feasible way to handle the noise‐signal relationship for more general topics of sensor data classification.

A lightweight multi‐branch network for low‐light image enhancement

AbstractThe generation process of low‐light images is fundamentally complicated due to their multi‐factorial degradation. Many previous methods favorably developed a multi‐branch network for impressive performance. However, the complex structures of multi‐branch networks always incur large computational cost and hinders their applicability in those resource‐limited settings. To alleviate these issues, in this letter a novel lightweight multi‐branch network (LMBN) for low‐light image enhancement is proposed. Specifically, a multi‐branch module and a joint loss are framed to surmount the complex degradation factors of low‐light images. Furthermore, the multi‐branch network is implemented by several heterogeneous and shallow encoder‐decoder modules. The elaborately designed lightweight model facilitates lower computational complexity but maintains impressive enhancement quality. The experimental results performed on three popularly benchmark databases demonstrate that the proposed LMBN shows the superior performance over other state‐of‐the‐art methods, indicating a perfect balance between the model's performance and the computational efficiency.

Pain Level and Pain-Related Behaviour Classification Using GRU-Based Sparsely-Connected RNNs

There is a growing body of studies on applying deep learning to biometrics analysis. Certain circumstances, however, could impair the objective measures and accuracy of the proposed biometric data analysis methods. For instance, people with chronic pain (CP) unconsciously adapt specific body movements to protect themselves from injury or additional pain. Because there is no dedicated benchmark database to analyse this correlation, we considered one of the specific circumstances that potentially influence a person's biometrics during daily activities in this study and classified pain level and pain-related behaviour in the EmoPain database. To achieve this, we proposed a sparsely-connected recurrent neural networks (s-RNNs) ensemble with the gated recurrent unit (GRU) that incorporates multiple autoencoders using a shared training framework. This architecture is fed by multidimensional data collected from inertial measurement unit (IMU) and surface electromyography (sEMG) sensors. Furthermore, to compensate for variations in the temporal dimension that may not be perfectly represented in the latent space of s-RNNs, we fused hand-crafted features derived from information-theoretic approaches with represented features in the shared hidden state. We conducted several experiments which indicate that the proposed method outperforms the state-of-the-art approaches in classifying both pain level and pain-related behaviour.

CNN — Forest Based Person Identification and Head Pose Estimation for AI Based Applications

Face recognition and head posture estimation have aroused a lot of academic interest recently since the inherent information improves the performance of face-related applications such as face alignment, augmented reality, healthcare applications, and emotion detection. The proposed work explores the challenges of identifying people and determining head posture. An analysis of the features produced at intermediate layers by limiting the number of kernals is performed and improved the performance of detecting the person. In addition, the learned features are sent to forest trees in order to determine the exact head attitude of the detected person. The proposed Forest CNN (FCNN) architecture is tested for head pose estimation methods on the Pointing 04 and Facepix benchmark databases.

An efficient lightweight network for single image super-resolution

The outstanding performance of convolutional neural networks (CNNs) shown in single image super-resolution (SISR) strongly depends on network’s depth, which hampers its application in low-power computing devices. In this paper, a lightweight and efficient network (LESR) is proposed for SISR by constructing the shallow feature extraction block (SFBlock), the cascaded sparse mask blocks (SMBlocks) and the feature fusion block (FFBlock). The SFBlock efficiently extracts global informative features from the original low resolution image using sparse self-attention, SMBlock skips the redundant computation in extracted features, and more meaningful information is distilled for the sequential reconstruction block by the FFBlock. In addition, a recently proposed activation function called ACON-C is used to replace the ReLU function to ease the training difficulty. Extensive experiments show that our proposed network performs better than most advanced lightweight SISR algorithms with comparable parameters and less FLOPs on benchmark database for ×2/3/4 SISR.

Lemniscate of Bernoulli's map quantifiers: innovative measures for EEG emotion recognition.

Thanks to the advent of affective computing, designing an automatic human emotion recognition system for clinical and non-clinical applications has attracted the attention of many researchers. Currently, multi-channel electroencephalogram (EEG)-based emotion recognition is a fundamental but challenging issue. This experiment envisioned developing a new scheme for automated EEG affect recognition. An innovative nonlinear feature engineering approach was presented based on Lemniscate of Bernoulli's Map (LBM), which belongs to the family of chaotic maps, in line with the EEG's nonlinear nature. As far as the authors know, LBM has not been utilized for biological signal analysis. Next, the map was characterized using several graphical indices. The feature vector was imposed on the feature selection algorithm while evaluating the role of the feature vector dimension on emotion recognition rates. Finally, the efficiency of the features on emotion recognition was appraised using two conventional classifiers and validated using the Database for Emotion Analysis using Physiological signals (DEAP) and SJTU Emotion EEG Dataset-IV (SEED-IV) benchmark databases. The experimental results showed a maximum accuracy of 92.16% for DEAP and 90.7% for SEED-IV. Achieving higher recognition rates compared to the state-of-art EEG emotion recognition systems suggest the proposed method based on LBM could have potential both in characterizing bio-signal dynamics and detecting affect-deficit disorders.

Effect of Hadamard multiplication on bloom filter and double bloom filter transformations

AbstractIris biometric is the most common and widely accepted biometric authentication method due to its high accuracy. The iris biometric template should be protected to overcome security and privacy attacks. The two standard iris biometric template protection methods are bloom filter and double bloom filter‐based feature transformations. In this article, we first permuted and then applied the p‐order Hadamard product on the iris template before the bloom filter and double bloom filter‐based transformation to enhance the security. The performance and security are measured using the CASIAv1 iris benchmark database to study the effect of Hadamard multiplication on bloom and double bloom filter‐based transformation. The Hadamard multiplication on the iris template increases the security of the transformed iris template and slightly degrades the matching performance.

Study on Using Linear Regression Modeling to Better Estimate Total Surgical Procedure Time

To make the most of the resources available in the operating room, careful planning of the available block time and the patient case sequences is required (OR). These planning methods will only help if the forecasts provided by TPT can be relied upon. This study makes use of linear regression models to improve TPT forecasts. These models are given estimated surgeon-controlled time (eSCT) in addition to other parameters of relevance. Surgical procedures at six Dutch institutions between 2012 and 2016 were evaluated using data from a benchmarking database. There were 79,983 contributors, which resulted in 199,772 hours of OR being reported. We analyzed the risk variables for TPT, which revealed that eSCT, patient age, surgical type, ASA physical status category, and anesthetic procedure were all contributors. During calculating the projected TPT for each file, the eSCT was multiplied by a fixed ratio model (discussed in further detail elsewhere) by a factor of 1.33. According to the findings of the research conducted by van Veen-Berkx et al., a conservative estimate of the period during which anesthesia is stable is around 33 percent of the SCT (ACT). In conclusion, we analyzed different linear regression models to predict TPT. These models included eSCT as one of the independent factors among the other covariates. In the end, each of the regression models was retested, this time to see whether or not it was possible to predict ACT reliably without using eSCT as a covariate (which leads to TPT by adding SCT). The most accurate projections of TPT were generated by a linear regression model that used eSCT, the kind of operation, the ASA classification, and the type of anesthesia. Compared to the fixed ratio model and the ACT prediction approach, the performance of this model was very satisfactory. Significant cost savings and increases in overall productivity may be realized if more accurate estimates were included in the planning and sequencing algorithms used to enhance operating room usage.

Training Spiking Neural Networks with Metaheuristic Algorithms

Taking inspiration from the brain, spiking neural networks (SNNs) have been proposed to understand and diminish the gap between machine learning and neuromorphic computing. Supervised learning is the most commonly used learning algorithm in traditional ANNs. However, directly training SNNs with backpropagation-based supervised learning methods is challenging due to the discontinuous and non-differentiable nature of the spiking neuron. To overcome these problems, this paper proposes a novel metaheuristic-based supervised learning method for SNNs by adapting the temporal error function. We investigated seven well-known metaheuristic algorithms called Harmony Search (HS), Cuckoo Search (CS), Differential Evolution (DE), Particle Swarm Optimization (PSO), Genetic Algorithm (GA), Artificial Bee Colony (ABC), and Grammatical Evolution (GE) as search methods for carrying out network training. Relative target firing times were used instead of fixed and predetermined ones, making the computation of the error function simpler. The performance of our proposed approach was evaluated using five benchmark databases collected in the UCI Machine Learning Repository. The experimental results showed that the proposed algorithm had a competitive advantage in solving the four classification benchmark datasets compared to the other experimental algorithms, with accuracy levels of 0.9858, 0.9768, 0.7752, and 0.6871 for iris, cancer, diabetes, and liver datasets, respectively. Among the seven metaheuristic algorithms, CS reported the best performance.

Dual Learning for Joint Facial Landmark Detection and Action Unit Recognition

Facial landmark detection and action unit~(AU) recognition are two essential tasks in facial analysis. Previous works rarely consider the relationship between these complementary tasks. In this paper, we introduce a novel multi-task dual learning framework to exploit the relationship between facial landmark detection and AU recognition while simultaneously addressing both tasks. When both tasks share middle-level features, common patterns can be exploited and middle- and high-level features can be used to perform facial landmark detection and AU recognition, respectively. In addition, a dual learning mechanism is designed to convert the predicted landmarks and AUs of the label space to the corresponding facial image of the image space, further exploring the strong correlations between the tasks. By jointly training the proposed method at both the feature and label levels, each task improves the other. Experiments on two benchmark databases demonstrate that the proposed method can leverage dependencies to boost the generalization of both tasks.

EEGformer: A transformer-based brain activity classification method using EEG signal.

The effective analysis methods for steady-state visual evoked potential (SSVEP) signals are critical in supporting an early diagnosis of glaucoma. Most efforts focused on adopting existing techniques to the SSVEPs-based brain-computer interface (BCI) task rather than proposing new ones specifically suited to the domain. Given that electroencephalogram (EEG) signals possess temporal, regional, and synchronous characteristics of brain activity, we proposed a transformer-based EEG analysis model known as EEGformer to capture the EEG characteristics in a unified manner. We adopted a one-dimensional convolution neural network (1DCNN) to automatically extract EEG-channel-wise features. The output was fed into the EEGformer, which is sequentially constructed using three components: regional, synchronous, and temporal transformers. In addition to using a large benchmark database (BETA) toward SSVEP-BCI application to validate model performance, we compared the EEGformer to current state-of-the-art deep learning models using two EEG datasets, which are obtained from our previous study: SJTU emotion EEG dataset (SEED) and a depressive EEG database (DepEEG). The experimental results show that the EEGformer achieves the best classification performance across the three EEG datasets, indicating that the rationality of our model architecture and learning EEG characteristics in a unified manner can improve model classification performance. EEGformer generalizes well to different EEG datasets, demonstrating our approach can be potentially suitable for providing accurate brain activity classification and being used in different application scenarios, such as SSVEP-based early glaucoma diagnosis, emotion recognition and depression discrimination.

The effect of node features on GCN-based brain network classification: an empirical study.

Brain functional network (BFN) analysis has become a popular technique for identifying neurological/mental diseases. Due to the fact that BFN is a graph, a graph convolutional network (GCN) can be naturally used in the classification of BFN. Different from traditional methods that directly use the adjacency matrices of BFNs to train a classifier, GCN requires an additional input-node features. To our best knowledge, however, there is no systematic study to analyze their influence on the performance of GCN-based brain disorder classification. Therefore, in this study, we conduct an empirical study on various node feature measures, including (1) original fMRI signals, (2) one-hot encoding, (3) node statistics, (4) node correlation, and (5) their combination. Experimental results on two benchmark databases show that different node feature inputs to GCN significantly affect the brain disease classification performance, and node correlation usually contributes higher accuracy compared to original signals and manually extracted statistical features.