Search

Space situational awareness (SSA) systems play a significant role in space navigation missions. One of the most essential tasks of this system is to recognize space objects such as spacecrafts and debris for various purposes including active debris removal, on-orbit servicing, and satellite formation. The complexity of object recognition in space is due to several sensing conditions, including the variety of object sizes with high contrast, low signal-to-noise ratio, noisy backgrounds, and several orbital scenarios. Existing methods have targeted the classification of images containing space objects with complex backgrounds using various convolutional neural networks. These methods sometimes lose attention on the objects in these images, which leads to misclassification and low accuracy. This paper proposes a decision fusion method that involves training an EfficientDet model with an EfficientNet-v2 backbone to detect space objects. Furthermore, the detected objects were augmented by blurring and by adding noise, and were then passed into the EfficientNet-B4 model for training. The decisions from both models were fused to find the final category among 11 categories. The experiments were conducted by utilizing a recently developed space object dataset (SPARK) generated from realistic space simulation environments. The dataset consists of 11 categories of objects with 150,000 RGB images and 150,000 depth images. The proposed object detection solution yielded superior performance and its feasibility for use in real-world SSA systems was demonstrated. Results show significant improvement in accuracy (94%), and performance metric (1.9223%) for object classification and in mean precision (78.45%) and mean recall (92.00%) for object detection. [ABSTRACT FROM AUTHOR]

Corrosion can initiate cracking that leads to structural integrity reduction. Quantitative corrosion assessment is challenging, and the modeling of corrosion-induced crack initiation is essential for model-based corrosion reliability analysis of various structures. This paper proposes a probabilistic computational analysis framework for corrosion-to-crack transitions by integrating a phase-field model with machine learning and uncertainty quantification. An electro-chemo-mechanical phase-field model is modified to predict pitting corrosion evolution, in which stress is properly coupled into the electrode chemical potential. A crack initiation criterion based on morphology is proposed to quantify the pit-to-cracking transition. A spatiotemporal surrogate modeling method is developed to facilitate this, consisting of a Convolution Neural Network (CNN) to map corrosion morphology to latent spaces, and a Gaussian Process regression model with a nonlinear autoregressive exogenous model (NARX) architecture for prediction of corrosion dynamics in the latent space over time. It enables the real-time prediction of corrosion morphology and crack initiation behaviors (whether, when, and where the corrosion damage triggers the crack initiation), and thus makes it possible for probabilistic analysis, with uncertainty quantified. Examples at various stress and corrosion conditions are presented to demonstrate the proposed computational framework. [ABSTRACT FROM AUTHOR]

The existing typical combined query image retrieval methods adopt Euclidean distance as sample distance measurement method, and the model trained by triple loss function blindly pursues absolute distance between samples, resulting in unsatisfactory image retrieval performance. Meanwhile, these methods singularly adopt Convolutional Neural Network (CNN) to extract reference image features. However, receptive field of convolution operation has the characteristics of locality, which is easy to cause the loss of edge feature information of reference images. In view of shortcomings of these methods, the following improvements are proposed in this paper: (1) We propose Triangle Area Triple Loss Function (TATLF), which adopts Triangle Area (TA) as measurement of sample distance. TA comprehensively considers the absolute distance and included angle between samples, so that the trained model has better retrieval performance; (2) We combine CNN with Transformer to simultaneously extract local and edge features of reference images, which can effectively reduce the loss of reference images information. Specifically, CNN is adopted to extract local feature information of reference images. Transformer is used to pay attention to the edge feature information of reference images. Extensive experiments on two public datasets, Fashion200k and MIT-States, confirm the excellent performance of our proposed method. [ABSTRACT FROM AUTHOR]

Facial expressions in non-human animals are closely linked to their internal affective states, with the majority of empirical work focusing on facial shape changes associated with pain. However, existing tools for facial expression analysis are prone to human subjectivity and bias, and in many cases also require special expertise and training. This paper presents the first comparative study of two different paths towards automatizing pain recognition in facial images of domestic short haired cats (n = 29), captured during ovariohysterectomy at different time points corresponding to varying intensities of pain. One approach is based on convolutional neural networks (ResNet50), while the other—on machine learning models based on geometric landmarks analysis inspired by species specific Facial Action Coding Systems (i.e. catFACS). Both types of approaches reach comparable accuracy of above 72%, indicating their potential usefulness as a basis for automating cat pain detection from images. [ABSTRACT FROM AUTHOR]

The detection of cancer stem-like cells (CSCs) is mainly based on molecular markers or functional tests giving a posteriori results. Therefore label-free and real-time detection of single CSCs remains a difficult challenge. The recent development of microfluidics has made it possible to perform high-throughput single cell imaging under controlled conditions and geometries. Such a throughput requires adapted image analysis pipelines while providing the necessary amount of data for the development of machine-learning algorithms. In this paper, we provide a data-driven study to assess the complexity of brightfield time-lapses to monitor the fate of isolated cancer stem-like cells in non-adherent conditions. We combined for the first time individual cell fate and cell state temporality analysis in a unique algorithm. We show that with our experimental system and on two different primary cell lines our optimized deep learning based algorithm outperforms classical computer vision and shallow learning-based algorithms in terms of accuracy while being faster than cutting-edge convolutional neural network (CNNs). With this study, we show that tailoring our deep learning-based algorithm to the image analysis problem yields better results than pre-trained models. As a result, such a rapid and accurate CNN is compatible with the rise of high-throughput data generation and opens the door to on-the-fly CSC fate analysis. [ABSTRACT FROM AUTHOR]

Dynamic complexity in brain functional connectivity has hindered the effective use of signal processing or machine learning methods to diagnose neurological disorders such as epilepsy. This paper proposed a new graph-generative neural network (GGN) model for the dynamic discovery of brain functional connectivity via deep analysis of scalp electroencephalogram (EEG) signals recorded from various regions of a patient's scalp. Brain functional connectivity graphs are generated for the extraction of spatial–temporal resolution of various onset epilepsy seizure patterns. Our supervised GGN model was substantiated by seizure detection and classification experiments. We train the GGN model using a clinically proven dataset of over 3047 epileptic seizure cases. The GGN model achieved a 91% accuracy in classifying seven types of epileptic seizure attacks, which outperformed the 65%, 74%, and 82% accuracy in using the convolutional neural network (CNN), graph neural networks (GNN), and transformer models, respectively. We present the GGN model architecture and operational steps to assist neuroscientists or brain specialists in using dynamic functional connectivity information to detect neurological disorders. Furthermore, we suggest to merge our spatial–temporal graph generator design in upgrading the conventional CNN and GNN models with dynamic convolutional kernels for accuracy enhancement. [ABSTRACT FROM AUTHOR]

Micro-expression is a kind of facial action that reflects the real emotional state of a person, and has high objectivity in emotion detection. Therefore, micro-expression recognition has become one of the research hotspots in the field of computer vision in recent years. Research with neural networks with convolutional structure is still one of the main methods of recognition. This method has the advantage of high operational efficiency and low computational complexity, but the disadvantage is its localization of feature extraction. In recent years, there are more and more plug-and-play self-attentive modules being used in convolutional neural networks to improve the ability of the model to extract global features of the samples. In this paper, we propose the ShuffleNet model combined with a miniature self-attentive module, which has only 1.53 million training parameters. First, the start frame and vertex frame of each sample will be taken out, and its TV-L1 optical flow features will be extracted. After that, the optical flow features are fed into the model for pre-training. Finally, the weights obtained from the pre-training are used as initialization weights for the model to train the complete micro-expression samples and classify them by the SVM classifier. To evaluate the effectiveness of the method, it was trained and tested on a composite dataset consisting of CASMEII, SMIC, and SAMM, and the model achieved competitive results compared to state-of-the-art methods through cross-validation of leave-one-out subjects. [ABSTRACT FROM AUTHOR]

The classification performance of all-optical Convolutional Neural Networks (CNNs) is greatly influenced by components' misalignment and translation of input images in the practical applications. In this paper, we propose a free-space all-optical CNN (named Trans-ONN) which accurately classifies translated images in the horizontal, vertical, or diagonal directions. Trans-ONN takes advantages of an optical motion pooling layer which provides the translation invariance property by implementing different optical masks in the Fourier plane for classifying translated test images. Moreover, to enhance the translation invariance property, global average pooling (GAP) is utilized in the Trans-ONN structure, rather than fully connected layers. The comparative studies confirm that taking advantage of vertical and horizontal masks along GAP operation provide the best translation invariance property, compared to the alternative network models, for classifying horizontally and vertically shifted test images up to 50 pixel shifts of Kaggle Cats and Dogs, CIFAR-10, and MNIST datasets, respectively. Also, adopting the diagonal mask along GAP operation achieves the best classification accuracy for classifying translated test images in the diagonal direction for large number of pixel shifts (i.e. more than 30 pixel shifts). It is worth mentioning that the proposed translation invariant networks are capable of classifying the translated test images not included in the training procedure. [ABSTRACT FROM AUTHOR]

Internet of Things (IoT) 's devices are ubiquitous and operate in a heterogonous environment with potential security breaches. IoT Operating Systems (IoT OSs) are the backbone software for running such devices. If IoT OSs are vulnerable to security breaches, higher-level security measures may not help. This paper aims to use Machine Learning (ML) to create a tool called iDetect for detecting vulnerabilities in C/C++ source code of IoT OSs. The source code for 16 releases of IoT OSs (RIOT, Contiki, FreeRTOS, Amazon FreeRTOS) and the Software Assurance Reference Dataset (SARD) were used to create a labeled dataset of vulnerable and benign code with the reference being the Common Weakness Enumeration (CWE) vulnerabilities present in IoT OSs. Studies showed that only a subset of CWEs is present in the C/C++ source code of low-end IoT OSs.The labeled dataset was used to train three ML models for vulnerability detection: Random Forest (RF), Convolutional Neural Network (CNN), and Recurrent Neural Network (RNN). The three models were used independently and RF; compared to CNN and RNN, gave the highest accuracy during the testing phase for binary and multiclass classification. RF was chosen as iDetect's ML classifier. Further evaluation was done on an unseen dataset of 322 code snippets taken from TinyOS. iDetect achieved a macro-averaged F1 score (mF1) of 98.5% and weighted-average F1 score (wF1) of 98% for multiclass classification, F1 score (F1) of 97.8% for binary classification, and superior results compared to all three Static Analysis Tools (SATs) used to collect the training dataset. [ABSTRACT FROM AUTHOR]

Increasing data infringement while transmission and storage have become an apprehension for the data owners. Even the digital images transmitted over the network or stored at servers are prone to unauthorized access. However, several image steganography techniques were proposed in the literature for hiding a secret image by embedding it into cover media. But the low embedding capacity and poor reconstruction quality of images are significant limitations of these techniques. To overcome these limitations, deep learning-based image steganography techniques are proposed in the literature. Convolutional neural network (CNN) based U-Net encoder has gained significant research attention in the literature. However, its performance efficacy as compared to other CNN based encoders like V-Net and U-Net++ is not implemented for image steganography. In this paper, V-Net and U-Net++ encoders are implemented for image steganography. A comparative performance assessment of U-Net, V-Net, and U-Net++ architectures are carried out. These architectures are employed to hide the secret image into the cover image. Further, a unique, robust, and standard decoder for all architectures is designed to extract the secret image from the cover image. Based on the experimental results, it is identified that U-Net architecture outperforms the other two architectures as it reports high embedding capacity and provides better quality stego and reconstructed secret images. [ABSTRACT FROM AUTHOR]

Due to the audio information of different types of vehicle models are distinct, the vehicle information can be identified by the audio signal of vehicle accurately. In real life, in order to determine the type of vehicle, we do not need to obtain the visual information of vehicles and just need to obtain the audio information. In this paper, we extract and stitching different features from different aspects: Mel frequency cepstrum coefficients in perceptual characteristics, pitch class profile in psychoacoustic characteristics and short-term energy in acoustic characteristics. In addition, we improve the neural networks classifier by fusing the LSTM unit into the convolutional neural networks. At last, we put the novel feature to the hybrid neural networks to recognize different vehicles. The results suggest the novel feature we proposed in this paper can increase the recognition rate by 7%; destroying the training data randomly by superimposing different kinds of noise can improve the anti-noise ability in our identification system; and LSTM has great advantages in modeling time series, adding LSTM to the networks can improve the recognition rate of 3.39%. [ABSTRACT FROM AUTHOR]

Breast cancer is a common malignancy and a leading cause of cancer-related deaths in women worldwide. Its early diagnosis can significantly reduce the morbidity and mortality rates in women. To this end, histopathological diagnosis is usually followed as the gold standard approach. However, this process is tedious, labor-intensive, and may be subject to inter-reader variability. Accordingly, an automatic diagnostic system can assist to improve the quality of diagnosis. This paper presents a deep learning approach to automatically classify hematoxylin-eosin-stained breast cancer microscopy images into normal tissue, benign lesion, in situ carcinoma, and invasive carcinoma using our collected dataset. Our proposed model exploited six intermediate layers of the Xception (Extreme Inception) network to retrieve robust and abstract features from input images. First, we optimized the proposed model on the original (unnormalized) dataset using 5-fold cross-validation. Then, we investigated its performance on four normalized datasets resulting from Reinhard, Ruifrok, Macenko, and Vahadane stain normalization. For original images, our proposed framework yielded an accuracy of 98% along with a kappa score of 0.969. Also, it achieved an average AUC-ROC score of 0.998 as well as a mean AUC-PR value of 0.995. Specifically, for in situ carcinoma and invasive carcinoma, it offered sensitivity of 96% and 99%, respectively. For normalized images, the proposed architecture performed better for Makenko normalization compared to the other three techniques. In this case, the proposed model achieved an accuracy of 97.79% together with a kappa score of 0.965. Also, it attained an average AUC-ROC score of 0.997 and a mean AUC-PR value of 0.991. Especially, for in situ carcinoma and invasive carcinoma, it offered sensitivity of 96% and 99%, respectively. These results demonstrate that our proposed model outperformed the baseline AlexNet as well as state-of-the-art VGG16, VGG19, Inception-v3, and Xception models with their default settings. Furthermore, it can be inferred that although stain normalization techniques offered competitive performance, they could not surpass the results of the original dataset. [ABSTRACT FROM AUTHOR]

Parameters such as gas content (GAS), porosity (PHI) and total organic carbon (TOC) are key parameters that reveal the shale gas geological "sweet spot" of reservoirs. However, the lack of a three-dimensional high-precision prediction method is not conducive to large-scale exploration of shale gas. Although the parameter prediction accuracy based on well logging data is relatively high, it is only a single point longitudinal feature. On the basis of prestack inversion of reservoir information such as P-wave velocity and density, high-precision and large-scale "sweet spot" spatial distribution predictions can be realized. Based on the fast growing and widely used deep learning methods, a one-dimensional convolutional neural network (1D-CNN) "sweet spot" parameter prediction method is proposed in this paper. First, intersection analysis is carried out for various well logging information to determine the sensitive parameters of geological "sweet spot". We propose a new standardized preprocessing method based on the characteristics of the well logging data. Then, a 1D-CNN framework is designed, which can meet the parameter prediction of both depth-domain well logging data and time-domain seismic data. Third, well logging data is used to train a high-precision and robust geological "sweet spot" prediction model. Finally, this method was applied to the WeiRong shale gas field in Sichuan Basin to achieve a high-precision prediction of geological "sweet spots" in the Wufeng–Longmaxi shale reservoir. [ABSTRACT FROM AUTHOR]

This paper introduces a deep learning based methodology for analyzing the self-assembled, fractal-like structures formed in evaporated droplets. To this end, an extensive image database of such structures of the plant extract Viscum album Quercus 10 - 3 was used, prepared by three different mixing procedures (turbulent, laminar, and diffusion based). The proposed pattern analysis approach is based on two stages: (1) automatic selection of patches that exhibit rich texture along the database; and (2) clustering of patches in accordance with prevalent texture by means of a Dense Convolutional Neural Network. The fractality of the patterns in each cluster is verified through Local Connected Fractal Dimension histograms. Experiments with Gray-Level Co-Occurrence matrices are performed to determine the benefit of the proposed approach in comparison with well established image analysis techniques. For the investigated plant extract, significant differences were found between the production modalities; whereas the patterns obtained by laminar flow showed the highest fractal structure, the patterns obtained by the application of turbulent mixture exhibited the lowest fractality. Our approach is the first to analyze, at the pure image level, the clustering properties of regions of interest within a database of evaporated droplets. This allows a greater description and differentiation of the patterns formed through different mixing procedures. [ABSTRACT FROM AUTHOR]

Activation functions in the neural network are responsible for 'firing' the nodes in it. In a deep neural network they 'activate' the features to reduce feature redundancy and learn the complex pattern by adding non-linearity in the network to learn task-specific goals. In this paper, we propose a simple and interesting activation function based on the combination of scaled gamma correction and hyperbolic tangent function, which we call Scaled Gamma Tanh (SGT) activation. The proposed activation function is applied in two steps, first is the calculation of gamma version as y = f(x) = axα for x < 0 and y = f(x) = bxβ for x ≥ 0, second is obtaining the squashed value as z = tanh(y). The variables a and b are user-defined constant values whereas α and β are channel-based learnable parameters. We analyzed the behavior of the proposed SGT activation function against other popular activation functions like ReLU, Leaky-ReLU, and tanh along with their role to confront vanishing/exploding gradient problems. For this, we implemented the SGT activation functions in a 3D Convolutional neural network (CNN) for the classification of magnetic resonance imaging (MRIs). More importantly to support our proposed idea we have presented a thorough analysis via histogram of inputs and outputs in activation layers along with weights/bias plot and t-SNE (t-Distributed Stochastic Neighbor Embedding) projection of fully connected layer for the trained CNN models. Our results in MRI classification show SGT outperforms standard ReLU and tanh activation in all cases i.e., final validation accuracy, final validation loss, test accuracy, Cohen's kappa score, and Precision. [ABSTRACT FROM AUTHOR]

Continuous technological growth and the corresponding environmental implications are triggering the enhancement of advanced environmental monitoring solutions, such as remote sensing. In this paper, we propose a new method for the spatial point patterns generation by classifying remote sensing images using convolutional neural network. To increase the accuracy, the training samples are extended by the suggested data augmentation scheme based on the similarities of images within the same part of the landscape for a limited observation time. The image patches are classified in accordance with the labels of previously classified images of the manually prepared training and test samples. This approach has improved the accuracy of image classification by 7% compared to current best practices of data augmentation. A set of image patch centers of a particular class is considered as a random point configuration, while the class labels are used as marks for every point. A marked point pattern is regarded as a combination of several subpoint patterns with the same qualitative marks. We analyze the bivariate point pattern to identify the relationships between points of different types using the features of a marked random point pattern. [ABSTRACT FROM AUTHOR]

Deep learning technology is changing the landscape of cybersecurity research, especially the study of large amounts of data. With the rapid growth in the number of malware, developing of an efficient and reliable method for classifying malware has become one of the research priorities. In this paper, a new method, BIR-CNN, is proposed to classify of Android malware. It combines convolution neural network (CNN) with batch normalization and inception-residual (BIR) network modules by using 347-dim network traffic features. CNN combines inception-residual modules with a convolution layer that can enhance the learning ability of the model. Batch Normalization can speed up the training process and avoid over-fitting of the model. Finally, experiments are conducted on the publicly available network traffic dataset CICAndMal2017 and compared with three traditional machine learning algorithms and CNN. The accuracy of BIR-CNN is 99.73% in binary classification (2-classifier). Moreover, the BIR-CNN can classify malware by its category (4-classifier) and malicious family (35-classifier), with a classification accuracy of 99.53% and 94.38%, respectively. The experimental results show that the proposed model is an effective method for Android malware classification, especially in malware category and family classifier. [ABSTRACT FROM AUTHOR]

The outdoor images captured in sand dust weather often suffer from poor contrast and color distortion, which seriously interfere with the performance of intelligent information processing systems. To solve the issues, a novel enhancement algorithm based on fusion strategy is proposed in this paper. It includes two components in sequence: sand removal via the improved Gaussian model-based color correction algorithm and dust elimination using the residual-based convolutional neural network (CNN). Theoretical analysis and experimental results show that compared with the prior sand dust image enhancement methods, the proposed fusion strategy can effectively correct the overall yellowing hue and remove the dust haze disturbance, which provides a constructive idea for the future development of sand dust image enhancement. [ABSTRACT FROM AUTHOR]

In the extensive search for new physics, the precise measurement of the Higgs boson continues to play an important role. To this end, machine learning techniques have been recently applied to processes like the Higgs production via vector-boson fusion. In this paper, we propose to use algorithms for learning to rank, i.e., to rank events into a sorting order, first signal, then background, instead of algorithms for the classification into two classes, for this task. The fact that training is then performed on pairwise comparisons of signal and background events can effectively increase the amount of training data due to the quadratic number of possible combinations. This makes it robust to unbalanced data set scenarios and can improve the overall performance compared to pointwise models like the state-of-the-art boosted decision tree approach. In this work we compare our pairwise neural network algorithm, which is a combination of a convolutional neural network and the DirectRanker, with convolutional neural networks, multilayer perceptrons or boosted decision trees, which are commonly used algorithms in multiple Higgs production channels. Furthermore, we use so-called transfer learning techniques to improve overall performance on different data types. [ABSTRACT FROM AUTHOR]

Accurate skull stripping facilitates following neuro-image analysis. For computer-aided methods, the presence of brain skull in structural magnetic resonance imaging (MRI) impacts brain tissue identification, which could result in serious misjudgments, specifically for patients with brain tumors. Though there are several existing works on skull stripping in literature, most of them either focus on healthy brain MRIs or only apply for a single image modality. These methods may be not optimal for multiparametric MRI scans. In the paper, we propose an ensemble neural network (EnNet), a 3D convolutional neural network (3DCNN) based method, for brain extraction on multiparametric MRI scans (mpMRIs). We comprehensively investigate the skull stripping performance by using the proposed method on a total of 15 image modality combinations. The comparison shows that utilizing all modalities provides the best performance on skull stripping. We have collected a retrospective dataset of 815 cases with/without glioblastoma multiforme (GBM) at the University of Pittsburgh Medical Center (UPMC) and The Cancer Imaging Archive (TCIA). The ground truths of the skull stripping are verified by at least one qualified radiologist. The quantitative evaluation gives an average dice score coefficient and Hausdorff distance at the 95th percentile, respectively. We also compare the performance to the state-of-the-art methods/tools. The proposed method offers the best performance. The contributions of the work have five folds: first, the proposed method is a fully automatic end-to-end for skull stripping using a 3D deep learning method. Second, it is applicable for mpMRIs and is also easy to customize for any MRI modality combination. Third, the proposed method not only works for healthy brain mpMRIs but also pre-/post-operative brain mpMRIs with GBM. Fourth, the proposed method handles multicenter data. Finally, to the best of our knowledge, we are the first group to quantitatively compare the skull stripping performance using different modalities. All code and pre-trained model are available at: https://github.com/plmoer/skull_stripping_code_SR. [ABSTRACT FROM AUTHOR]

Based on physics-informed deep learning method, the deep learning model is proposed for thermal fluid fields reconstruction. This method applied fully-connected layers to establish the mapping function from design variables and space coordinates to physical fields of interest, and then the performance characteristics Nusselt number Nu and Fanning friction factor f can be calculated from the reconstructed fields. Compared with reconstruction model based on convolutional neural network, the improved model shows no constrains on mesh generation and it improves the physical interpretability by introducing conservation laws in loss functions. To validate this method, the forced convection of the water-Al2O3 nanofluids is utilized to construct training dataset. As shown in this paper, this deep neural network can reconstruct the physical fields and consequently the performance characteristics accurately. In the comparisons with other classical machine learning methods, our reconstruction model is superior for predicting performance characteristics. In addition to the effect of training size on prediction power, the extrapolation performance (an important but rarely investigated issue) for important design parameters are also explored on unseen testing datasets. [ABSTRACT FROM AUTHOR]

Accurate skull stripping facilitates following neuro-image analysis. For computer-aided methods, the presence of brain skull in structural magnetic resonance imaging (MRI) impacts brain tissue identification, which could result in serious misjudgments, specifically for patients with brain tumors. Though there are several existing works on skull stripping in literature, most of them either focus on healthy brain MRIs or only apply for a single image modality. These methods may be not optimal for multiparametric MRI scans. In the paper, we propose an ensemble neural network (EnNet), a 3D convolutional neural network (3DCNN) based method, for brain extraction on multiparametric MRI scans (mpMRIs). We comprehensively investigate the skull stripping performance by using the proposed method on a total of 15 image modality combinations. The comparison shows that utilizing all modalities provides the best performance on skull stripping. We have collected a retrospective dataset of 815 cases with/without glioblastoma multiforme (GBM) at the University of Pittsburgh Medical Center (UPMC) and The Cancer Imaging Archive (TCIA). The ground truths of the skull stripping are verified by at least one qualified radiologist. The quantitative evaluation gives an average dice score coefficient and Hausdorff distance at the 95th percentile, respectively. We also compare the performance to the state-of-the-art methods/tools. The proposed method offers the best performance. The contributions of the work have five folds: first, the proposed method is a fully automatic end-to-end for skull stripping using a 3D deep learning method. Second, it is applicable for mpMRIs and is also easy to customize for any MRI modality combination. Third, the proposed method not only works for healthy brain mpMRIs but also pre-/post-operative brain mpMRIs with GBM. Fourth, the proposed method handles multicenter data. Finally, to the best of our knowledge, we are the first group to quantitatively compare the skull stripping performance using different modalities. All code and pre-trained model are available at: https://github.com/plmoer/skull_stripping_code_SR. [ABSTRACT FROM AUTHOR]

Factories swiftly and precisely grasp the real-time data of the production instrumentation, which is the foundation for the development and progress of industrial intelligence in industrial production. Weather, light, angle, and other unknown circumstances, on the other hand, impair the image quality of meter dials in natural environments, resulting in poor dial image quality. The remote meter reading system has trouble recognizing dial pictures in extreme settings, challenging it to meet industrial production demands. This paper provides multiple attention and encoder–decoder-based gas meter recognition networks (MAEDR) for this problem. First, from the acquired dial photos, the dial images with extreme conditions such as overexposure, artifacts, blurring, incomplete display of characters, and occlusion are chosen to generate the gas meter dataset. Then, a new character recognition network is proposed utilizing multiple attention and an encoder–decoder structure. Convolutional neural networks (CNN) extract visual features from dial images, encode visual features employing multi-head self-attention and position information, and facilitate feature alignment using the connectionist temporal classification (CTC) method. A novel two-step attention decoder is presented to improve the accuracy of recognition results. convolutional block attention module (CBAM) reweights the visual features from the CNN and the semantic features computed by the encoder to improve model performance; long short-term memory attention (LSTM attention) focuses on the relationship between feature sequences. According to experimental data, our system can effectively and efficiently identify industrial gas meter picture digits with 91.1% identification accuracy, faster inference speed, and higher accuracy than standard algorithms. The accuracy and practicality of the recognition can fulfill the needs of instrument data detection and recognition in industrial production, and it has a wide range of applications. [ABSTRACT FROM AUTHOR]

Due to the effect of emotions on interactions, interpretations, and decisions, automatic detection and analysis of human emotions based on EEG signals has an important role in the treatment of psychiatric diseases. However, the low spatial resolution of EEG recorders poses a challenge. In order to overcome this problem, in this paper we model each emotion by mapping from scalp sensors to brain sources using Bernoulli–Laplace-based Bayesian model. The standard low-resolution electromagnetic tomography (sLORETA) method is used to initialize the source signals in this algorithm. Finally, a dynamic graph convolutional neural network (DGCNN) is used to classify emotional EEG in which the sources of the proposed localization model are considered as the underlying graph nodes. In the proposed method, the relationships between the EEG source signals are encoded in the DGCNN adjacency matrix. Experiments on our EEG dataset recorded at the Brain-Computer Interface Research Laboratory, University of Tabriz as well as publicly available SEED and DEAP datasets show that brain source modeling by the proposed algorithm significantly improves the accuracy of emotion recognition, such that it achieve a classification accuracy of 99.25% during the classification of the two classes of positive and negative emotions. These results represent an absolute 1–2% improvement in terms of classification accuracy over subject-dependent and subject-independent scenarios over the existing approaches. [ABSTRACT FROM AUTHOR]

Facial expressions in non-human animals are closely linked to their internal affective states, with the majority of empirical work focusing on facial shape changes associated with pain. However, existing tools for facial expression analysis are prone to human subjectivity and bias, and in many cases also require special expertise and training. This paper presents the first comparative study of two different paths towards automatizing pain recognition in facial images of domestic short haired cats (n = 29), captured during ovariohysterectomy at different time points corresponding to varying intensities of pain. One approach is based on convolutional neural networks (ResNet50), while the other—on machine learning models based on geometric landmarks analysis inspired by species specific Facial Action Coding Systems (i.e. catFACS). Both types of approaches reach comparable accuracy of above 72%, indicating their potential usefulness as a basis for automating cat pain detection from images. [ABSTRACT FROM AUTHOR]

Monitoring the conditions of machines is vital in the manufacturing industry. Early detection of faulty components in machines for stopping and repairing the failed components can minimize the downtime of the machine. In this article, we present a method for detecting failures in drill machines using drill sounds in Valmet AB, a company in Sundsvall, Sweden that supplies equipment and processes for the production of pulp, paper, and biofuels. The drill dataset includes two classes: anomalous sounds and normal sounds. Detecting drill failure effectively remains a challenge due to the following reasons. The waveform of drill sound is complex and short for detection. Furthermore, in realistic soundscapes, both sounds and noise exist simultaneously. Besides, the balanced dataset is small to apply state-of-the-art deep learning techniques. Due to these aforementioned difficulties, sound augmentation methods were applied to increase the number of sounds in the dataset. In this study, a convolutional neural network (CNN) was combined with a long-short-term memory (LSTM) to extract features from log-Mel spectrograms and to learn global representations of two classes. A leaky rectified linear unit (Leaky ReLU) was utilized as the activation function for the proposed CNN instead of the ReLU. Moreover, an attention mechanism was deployed at the frame level after the LSTM layer to pay attention to the anomaly in sounds. As a result, the proposed method reached an overall accuracy of 92.62% to classify two classes of machine sounds on Valmet's dataset. In addition, an extensive experiment on another drilling dataset with short sounds yielded 97.47% accuracy. With multiple classes and long-duration sounds, an experiment utilizing the publicly available UrbanSound8K dataset obtains 91.45%. Extensive experiments on our dataset as well as publicly available datasets confirm the efficacy and robustness of our proposed method. For reproducing and deploying the proposed system, an open-source repository is publicly available at https://github.com/thanhtran1965/DrillFailureDetection_SciRep2022. [ABSTRACT FROM AUTHOR]

With the intensification of ecosystem damage, birds have become the symbolic species of the ecosystem. Ornithology with interdisciplinary technical research plays a great significance for protecting birds and evaluating ecosystem quality. Deep learning shows great progress for birdsongs recognition. However, as the number of network layers increases in traditional CNN, semantic information gradually becomes richer and detailed information disappears. Secondly, the global information carried by the entire input may be lost in convolution, pooling, or other operations, and these problems will weaken the performance of classification. In order to solve such problems, based on the feature spectrogram from the wavelet transform for the birdsongs, this paper explored the multi-scale convolution neural network (MSCNN) and proposed an ensemble multi-scale convolution neural network (EMSCNN) classification framework. The experiments compared the MSCNN and EMSCNN models with other CNN models including LeNet, VGG16, ResNet101, MobileNetV2, EfficientNetB7, Darknet53 and SPP-net. The results showed that the MSCNN model achieved an accuracy of 89.61%, and EMSCNN achieved an accuracy of 91.49%. In the experiments on the recognition of 30 species of birds, our models effectively improved the classification effect with high stability and efficiency, indicating that the models have better generalization ability and are suitable for birdsongs species recognition. It provides methodological and technical scheme reference for bird classification research. [ABSTRACT FROM AUTHOR]

Boosted by mobile communication technologies, Human Activity Recognition (HAR) based on smartphones has attracted more and more attentions of researchers. One of the main challenges is the classification time and accuracy in processing long-time dependent sequence samples with noisy or missed data. In this paper, a 1-D Convolution Neural Network (CNN)-based bi-directional Long Short-Term Memory (LSTM) parallel model with attention mechanism (ConvBLSTM-PMwA) is proposed. The original features of sensors are segmented into sub-segments by well-designed equal time step sliding window, and fed into 1-D CNN-based bi-directional LSTM parallel layer to accelerate feature extraction with noisy and missed data. The weights of extracted features are redistributed by attention mechanism and integrated into complete features. At last, the final classification results are obtained with the full connection layer. The performance is evaluated on public UCI and WISDM HAR datasets. The results show that the ConvBLSTM-PMwA model performs better than the existing CNN and RNN models in both classification accuracy (96.71%) and computational time complexity (1.1 times faster at least), even if facing HAR data with noise. [ABSTRACT FROM AUTHOR]

Deep hashing method is widely applied in the field of image retrieval because of its advantages of low storage consumption and fast retrieval speed. There is a defect of insufficiency feature extraction when existing deep hashing method uses the convolutional neural network (CNN) to extract images semantic features. Some studies propose to add channel-based or spatial-based attention modules. However, embedding these modules into the network can increase the complexity of model and lead to over fitting in the training process. In this study, a novel deep parameter-free attention hashing (DPFAH) is proposed to solve these problems, that designs a parameter-free attention (PFA) module in ResNet18 network. PFA is a lightweight module that defines an energy function to measure the importance of each neuron and infers 3-D attention weights for feature map in a layer. A fast closed-form solution for this energy function proves that the PFA module does not add any parameters to the network. Otherwise, this paper designs a novel hashing framework that includes the hash codes learning branch and the classification branch to explore more label information. The like-binary codes are constrained by a regulation term to reduce the quantization error in the continuous relaxation. Experiments on CIFAR-10, NUS-WIDE and Imagenet-100 show that DPFAH method achieves better performance. [ABSTRACT FROM AUTHOR]

Emotion recognition is defined as identifying human emotion and is directly related to different fields such as human–computer interfaces, human emotional processing, irrational analysis, medical diagnostics, data-driven animation, human–robot communication, and many more. This paper proposes a new facial emotional recognition model using a convolutional neural network. Our proposed model, "ConvNet", detects seven specific emotions from image data including anger, disgust, fear, happiness, neutrality, sadness, and surprise. The features extracted by the Local Binary Pattern (LBP), region based Oriented FAST and rotated BRIEF (ORB) and Convolutional Neural network (CNN) from facial expressions images were fused to develop the classification model through training by our proposed CNN model (ConvNet). Our method can converge quickly and achieves good performance which the authors can develop a real-time schema that can easily fit the model and sense emotions. Furthermore, this study focuses on the mental or emotional stuff of a man or woman using the behavioral aspects. To complete the training of the CNN network model, we use the FER2013 databases at first, and then apply the generalization techniques to the JAFFE and CK+ datasets respectively in the testing stage to evaluate the performance of the model. In the generalization approach on the JAFFE dataset, we get a 92.05% accuracy, while on the CK+ dataset, we acquire a 98.13% accuracy which achieve the best performance among existing methods. We also test the system's success by identifying facial expressions in real-time. ConvNet consists of four layers of convolution together with two fully connected layers. The experimental results show that the ConvNet is able to achieve 96% training accuracy which is much better than current existing models. However, when compared to other validation methods, the suggested technique was more accurate. ConvNet also achieved validation accuracy of 91.01% for the FER2013 dataset. We also made all the materials publicly accessible for the research community at: https://github.com/Tanoy004/Emotion-recognition-through-CNN. [ABSTRACT FROM AUTHOR]

Recently, with the construction of smart city, the research on environmental sound classification (ESC) has attracted the attention of academia and industry. The development of convolutional neural network (CNN) makes the accuracy of ESC reach a higher level, but the accuracy improvement brought by CNN is often accompanied by the deepening of network layers, which leads to the rapid growth of parameters and floating-point operations (FLOPs). Therefore, it is difficult to transplant CNN model to embedded devices, and the classification speed is also difficult to accept. In order to reduce the hardware requirements of running CNN and improve the speed of ESC, this paper proposes a resource adaptive convolutional neural network (RACNN). RACNN uses a novel resource adaptive convolutional (RAC) module, which can generate the same number of feature maps as conventional convolution operations more cheaply, and extract the time and frequency features of audio efficiently. The RAC block based on the RAC module is designed to build the lightweight RACNN model, and the RAC module can also be used to upgrade the existing CNN model. Experiments based on public datasets show that RACNN achieves higher performance than the state-of-the-art methods with lower computational complexity. [ABSTRACT FROM AUTHOR]

Head kinematics information is important as it is used to measure brain injury risk. Currently, head kinematics are measured using wearable devices or instrumentation mounted on the head. This paper evaluates the deep learning approach in predicting time history of head angular kinematics directly from videos without any instrumentation. To prove the concept, a deep learning model was developed for predicting time history of head angular velocities using finite element (FE) based crash simulation videos. This FE dataset was split into training, validation, and test datasets. A combined convolutional neural network and recurrent neural network based deep learning model was developed using the training and validations sets. The test (unseen) dataset was used to evaluate the predictive capability of the deep learning model. On the test dataset, correlation coefficient obtained between the actual and predicted peak angular velocities was 0.73, 0.85, and 0.92 for X, Y, and Z components respectively. [ABSTRACT FROM AUTHOR]

Head kinematics information is important as it is used to measure brain injury risk. Currently, head kinematics are measured using wearable devices or instrumentation mounted on the head. This paper evaluates the deep learning approach in predicting time history of head angular kinematics directly from videos without any instrumentation. To prove the concept, a deep learning model was developed for predicting time history of head angular velocities using finite element (FE) based crash simulation videos. This FE dataset was split into training, validation, and test datasets. A combined convolutional neural network and recurrent neural network based deep learning model was developed using the training and validations sets. The test (unseen) dataset was used to evaluate the predictive capability of the deep learning model. On the test dataset, correlation coefficient obtained between the actual and predicted peak angular velocities was 0.73, 0.85, and 0.92 for X, Y, and Z components respectively. [ABSTRACT FROM AUTHOR]

Accurate glioma subtype classification is critical for the treatment management of patients with brain tumors. Developing an automatically computer-aided algorithm for glioma subtype classification is challenging due to many factors. One of the difficulties is the label constraint. Specifically, each case is simply labeled the glioma subtype without precise annotations of lesion regions information. In this paper, we propose a novel hybrid fully convolutional neural network (CNN)-based method for glioma subtype classification using both whole slide imaging (WSI) and multiparametric magnetic resonance imagings (mpMRIs). It is comprised of two methods: a WSI-based method and a mpMRIs-based method. For the WSI-based method, we categorize the glioma subtype using a 2D CNN on WSIs. To overcome the label constraint issue, we extract the truly representative patches for the glioma subtype classification in a weakly supervised fashion. For the mpMRIs-based method, we develop a 3D CNN-based method by analyzing the mpMRIs. The mpMRIs-based method consists of brain tumor segmentation and classification. Finally, to enhance the robustness of the predictions, we fuse the WSI-based and mpMRIs-based results guided by a confidence index. The experimental results on the validation dataset in the competition of CPM-RadPath 2020 show the comprehensive judgments from both two modalities can achieve better performance than the ones by solely using WSI or mpMRIs. Furthermore, our result using the proposed method ranks the third place in the CPM-RadPath 2020 in the testing phase. The proposed method demonstrates a competitive performance, which is creditable to the success of weakly supervised approach and the strategy of label agreement from multi-modality data. [ABSTRACT FROM AUTHOR]

Videos, especially short videos, have become an increasingly important source of information in these years. However, many videos spread on video sharing platforms are misleading, which have negative social impacts. Therefore, it is necessary to find methods to automatically identify misleading videos. In this paper, three categories of features (content features, uploader features and environment features) are proposed to construct a convolutional neural network (CNN) for misleading video detection. The experiment showed that all the three proposed categories of features play a vital role in detecting misleading videos. Our proposed approach that combines three categories of features achieved the best performance with the accuracy of 0.90 and the F1 score of 0.89. It also outperformed other baselines such as SVM, k-NN, decision tree and random forest models by more than 22%. [ABSTRACT FROM AUTHOR]

The robotic manipulation of a heavy industrial cable is challenging to model and control because of the high number of degrees of freedom and the rigid-flexible coupling dynamics. In this paper, we report the development of modeling the cable effect and control methodology for robotic cable manipulation. Our cable effect model is based on the 2D convolutional neural network, which is a deep learning-based method uses the effective cable representation method to achieve the accurate, generalizable, and efficient estimation of the cable coupling forces and torques. Practical problems such as the measurement limits and time efficiency are considered in our method for real applications. With these approaches, we are the first to solve the problem of dynamic payload effect caused by heavy industrial cables in experimental cases. The used control methodology combines the active disturbance rejection control framework with the sliding mode control method, which can acquire promising tracking performance. We integrate our cable effect model into the control scheme, and demonstrate it satisfies the high-quality robotic manipulation of heavy cables. The performance of the proposed method is assessed with both a simulated system and real robot system. The results show that our method can estimate the cable coupling effect with over 85% accuracy and accomplish manipulation with a positioning error less than 0.01 mm. This reveals that our method is promising for robotic manipulation of heavy industrial cables and can accomplish the challenging cable insertion task. [ABSTRACT FROM AUTHOR]

Recognition of space objects including spacecraft and debris is one of the main components in the space situational awareness (SSA) system. Various tasks such as satellite formation, on-orbit servicing, and active debris removal require object recognition to be done perfectly. The recognition task in actual space imagery is highly complex because the sensing conditions are largely diverse. The conditions include various backgrounds affected by noise, several orbital scenarios, high contrast, low signal-to-noise ratio, and various object sizes. To address the problem of space recognition, this paper proposes a multi-modal learning solution using various deep learning models. To extract features from RGB images that have spacecraft and debris, various convolutional neural network (CNN) based models such as ResNet, EfficientNet, and DenseNet were explored. Furthermore, RGB based vision transformer was demonstrated. Additionally, End-to-End CNN was used for classification of depth images. The final decision of the proposed solution combines the two decisions from RGB based and Depth-based models. The experiments were carried out using a novel dataset called SPARK which was generated under a realistic space simulation environment. The dataset includes various images with eleven categories, and it is divided into 150 k of RGB images and 150 k of depth images. The proposed combination of RGB based vision transformer and Depth-based End-to-End CNN showed higher performance and better results in terms of accuracy (85%), precision (86%), recall (85%), and F1 score (84%). Therefore, the proposed multi-modal deep learning is a good feasible solution to be utilized in real tasks of SSA system. [ABSTRACT FROM AUTHOR]

Fully automated and volumetric segmentation of critical tumors may play a crucial role in diagnosis and surgical planning. One of the most challenging tumor segmentation tasks is localization of pancreatic ductal adenocarcinoma (PDAC). Exclusive application of conventional methods does not appear promising. Deep learning approaches has achieved great success in the computer aided diagnosis, especially in biomedical image segmentation. This paper introduces a framework based on convolutional neural network (CNN) for segmentation of PDAC mass and surrounding vessels in CT images by incorporating powerful classic features, as well. First, a 3D-CNN architecture is used to localize the pancreas region from the whole CT volume using 3D Local Binary Pattern (LBP) map of the original image. Segmentation of PDAC mass is subsequently performed using 2D attention U-Net and Texture Attention U-Net (TAU-Net). TAU-Net is introduced by fusion of dense Scale-Invariant Feature Transform (SIFT) and LBP descriptors into the attention U-Net. An ensemble model is then used to cumulate the advantages of both networks using a 3D-CNN. In addition, to reduce the effects of imbalanced data, a multi-objective loss function is proposed as a weighted combination of three classic losses including Generalized Dice Loss (GDL), Weighted Pixel-Wise Cross Entropy loss (WPCE) and boundary loss. Due to insufficient sample size for vessel segmentation, we used the above-mentioned pre-trained networks and fine-tuned them. Experimental results show that the proposed method improves the Dice score for PDAC mass segmentation in portal-venous phase by 7.52% compared to state-of-the-art methods in term of DSC. Besides, three dimensional visualization of the tumor and surrounding vessels can facilitate the evaluation of PDAC treatment response. [ABSTRACT FROM AUTHOR]

The increasing use of nuclear technology in various fields makes it necessary to provide the required safety to work with this industry. Gamma source is one of the most widely used sources in industry and medicine. Finding a lost gamma source in a gamma irradiation room without human presence is challenging due to the particular arrangements and barriers in the room for radiation shielding and requires an efficient and robust method. In this paper, locating and routing the lost gamma source in the gamma irradiation room containing radiation blocking barriers are done simultaneously by using two methods, convolutional neural network (CNN) and Q-learning, which are powerful algorithms for deep learning and machine learning. Environment simulation with gamma source was performed using Geant4 simulation. The results show that by combining these two methods in geometries with radiation blocking barriers, in addition to locating with 90% accuracy, routing can also be performed. Although the presence of thick barriers in the room reduces the accuracy, increases the time required to finding the lost gamma source or the inefficiency of other methods, nevertheless, the results show that combination of CNN and Q-learning reduces the time and greatly increases the accuracy. [ABSTRACT FROM AUTHOR]

In this study, we investigate perspectives for thermal tomography based on planar infrared thermal images. Volumetric reconstruction of temperature distribution inside an object is hardly applicable in a way similar to ionizing-radiation-based modalities due to its non-penetrating character. Here, we aim at employing the autoencoder deep neural network to collect knowledge on the single-source heat transfer model. For that purpose, we prepare a series of synthetic 3D models of a cylindrical phantom with assumed thermal properties with various heat source locations, captured at different times. A set of planar thermal images taken around the model is subjected to initial backprojection reconstruction, then passed to the deep model. This paper reports the training and testing results in terms of five metrics assessing spatial similarity between volumetric models, signal-to-noise ratio, or heat source location accuracy. We also evaluate the assumptions of the synthetic model with an experiment involving thermal imaging of a real object (pork) and a single heat source. For validation, we investigate objects with multiple heat sources of a random location and temperature. Our results show the capability of a deep model to reconstruct the temperature distribution inside the object. [ABSTRACT FROM AUTHOR]

Pulmonary nodules are the main manifestation of early lung cancer. Therefore, accurate detection of nodules in CT images is vital for lung cancer diagnosis. A 3D automatic detection system of pulmonary nodules based on multi-scale attention networks is proposed in this paper to use multi-scale features of nodules and avoid network over-fitting problems. The system consists of two parts, nodule candidate detection (determining the locations of candidate nodules), false positive reduction (minimizing the number of false positive nodules). Specifically, with Res2Net structure, using pre-activation operation and convolutional quadruplet attention module, the 3D multi-scale attention block is designed. It makes full use of multi-scale information of pulmonary nodules by extracting multi-scale features at a granular level and alleviates over-fitting by pre-activation. The U-Net-like encoder-decoder structure is combined with multi-scale attention blocks as the backbone network of Faster R-CNN for detection of candidate nodules. Then a 3D deep convolutional neural network based on multi-scale attention blocks is designed for false positive reduction. The extensive experiments on LUNA16 and TianChi competition datasets demonstrate that the proposed approach can effectively improve the detection sensitivity and control the number of false positive nodules, which has clinical application value. [ABSTRACT FROM AUTHOR]

Supervised machine learning is a traditionally remaining useful life (RUL) estimation tool, which requires a lot of prior knowledge. For the situation lacking labeled data, supervised methods are invalid for the issue of domain shift in data distribution. In this paper, a adversarial-based domain adaptation (ADA) architecture with convolution neural networks (CNN) for RUL estimation of bearings under different conditions and platforms, referred to as ADACNN, is proposed. Specifically, ADACNN is trained in source labeled data and fine-tunes to similar target unlabeled data via an adversarial training and parameters shared mechanism. Besides a feature extractor and source domain regressive predictor, ADACNN also includes a domain classifier that tries to guide feature extractor find some domain-invariant features, which differents with traditional methods and belongs to a unsupervised learning in target domain, which has potential application value and far-reaching significance in academia. In addition, according to different first predictive time (FPT) detection mechanisms, we also explores the impact of different FPT detection mechanisms on RUL estimation performance. Finally, according to extensive experiments, the results of RUL estimation of bearing in cross-condition and cross-platform prove that ADACNN architecture has satisfactory generalization performance and great practical value in industry. [ABSTRACT FROM AUTHOR]

Vision loss happens due to diabetic retinopathy (DR) in severe stages. Thus, an automatic detection method applied to diagnose DR in an earlier phase may help medical doctors to make better decisions. DR is considered one of the main risks, leading to blindness. Computer-Aided Diagnosis systems play an essential role in detecting features in fundus images. Fundus images may include blood vessels, exudates, micro-aneurysm, hemorrhages, and neovascularization. In this paper, our model combines automatic detection for the diabetic retinopathy classification with localization methods depending on weakly-supervised learning. The model has four stages; in stage one, various preprocessing techniques are applied to smooth the data set. In stage two, the network had gotten deeply to the optic disk segment for eliminating any exudate's false prediction because the exudates had the same color pixel as the optic disk. In stage three, the network is fed through training data to classify each label. Finally, the layers of the convolution neural network are re-edited, and used to localize the impact of DR on the patient's eye. The framework tackles the matching technique between two essential concepts where the classification problem depends on the supervised learning method. While the localization problem was obtained by the weakly supervised method. An additional layer known as weakly supervised sensitive heat map (WSSH) was added to detect the ROI of the lesion at a test accuracy of 98.65%, while comparing with Class Activation Map that involved weakly supervised technology achieved 0.954. The main purpose is to learn a representation that collect the central localization of discriminative features in a retina image. CNN-WSSH model is able to highlight decisive features in a single forward pass for getting the best detection of lesions. [ABSTRACT FROM AUTHOR]

This paper proposes a method that automatically measures non-invasive blood pressure (BP) based on an auscultatory approach using Korotkoff sounds (K-sounds). There have been methods utilizing K-sounds that were more accurate in general than those using cuff pressure signals only under well-controlled environments, but most were vulnerable to the measurement conditions and to external noise because blood pressure is simply determined based on threshold values in the sound signal. The proposed method enables robust and precise BP measurements by evaluating the probability that each sound pulse is an audible K-sound based on a deep learning using a convolutional neural network (CNN). Instead of classifying sound pulses into two categories, audible K-sounds and others, the proposed CNN model outputs probability values. These values in a Korotkoff cycle are arranged in time order, and the blood pressure is determined. The proposed method was tested with a dataset acquired in practice that occasionally contains considerable noise, which can degrade the performance of the threshold-based methods. The results demonstrate that the proposed method outperforms a previously reported CNN-based classification method using K-sounds. With larger amounts of various types of data, the proposed method can potentially achieve more precise and robust results. [ABSTRACT FROM AUTHOR]

Recently, several convolutional neural networks have been proposed not only for 2D images, but also for 3D and 4D volume segmentation. Nevertheless, due to the large data size of the latter, acquiring a sufficient amount of training annotations is much more strenuous than in 2D images. For 4D time-series tomograms, this is usually handled by segmenting the constituent tomograms independently through time with 3D convolutional neural networks. Inter-volume information is therefore not utilized, potentially leading to temporal incoherence. In this paper, we attempt to resolve this by proposing two hidden Markov model variants that refine 4D segmentation labels made by 3D convolutional neural networks working on each time point. Our models utilize not only inter-volume information, but also the prediction confidence generated by the 3D segmentation convolutional neural networks themselves. To the best of our knowledge, this is the first attempt to refine 4D segmentations made by 3D convolutional neural networks using hidden Markov models. During experiments we test our models, qualitatively, quantitatively and behaviourally, using prespecified segmentations. We demonstrate in the domain of time series tomograms which are typically undersampled to allow more frequent capture; a particularly challenging problem. Finally, our dataset and code is publicly available. [ABSTRACT FROM AUTHOR]

Environmental sound classification is one of the important issues in the audio recognition field. Compared with structured sounds such as speech and music, the time–frequency structure of environmental sounds is more complicated. In order to learn time and frequency features from Log-Mel spectrogram more effectively, a temporal-frequency attention based convolutional neural network model (TFCNN) is proposed in this paper. Firstly, an experiment that is used as motivation in proposed method is designed to verify the effect of a specific frequency band in the spectrogram on model classification. Secondly, two new attention mechanisms, temporal attention mechanism and frequency attention mechanism, are proposed. These mechanisms can focus on key frequency bands and semantic related time frames on the spectrogram to reduce the influence of background noise and irrelevant frequency bands. Then, a feature information complementarity is formed by combining these mechanisms to more accurately capture the critical time–frequency features. In such a way, the representation ability of the network model can be greatly improved. Finally, experiments on two public data sets, UrbanSound 8 K and ESC-50, demonstrate the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Triple negative (TN) breast cancer is a subtype of breast cancer which is difficult for early detection and the prognosis is poor. In this paper, 910 benign and 934 malignant (110 TN and 824 NTN) B-mode breast ultrasound images were collected. A Resnet50 deep convolutional neural network was fine-tuned. The results showed that the averaged area under the receiver operating characteristic curve (AUC) of discriminating malignant from benign ones were 0.9789 (benign vs. TN), 0.9689 (benign vs. NTN). To discriminate TN from NTN breast cancer, the AUC was 0.9000, the accuracy was 88.89%, the sensitivity was 87.5%, and the specificity was 90.00%. It showed that the computer-aided system based on DCNN is expected to be a promising noninvasive clinical tool for ultrasound diagnosis of TN breast cancer. [ABSTRACT FROM AUTHOR]

An artificial neural network consists of neurons and synapses. Neuron gives output based on its input according to non-linear activation functions such as the Sigmoid, Hyperbolic Tangent (Tanh), or Rectified Linear Unit (ReLU) functions, etc.. Synapses connect the neuron outputs to their inputs with tunable real-valued weights. The most resource-demanding operations in realizing such neural networks are the multiplication and accumulate (MAC) operations that compute the dot product between real-valued outputs from neurons and the synapses weights. The efficiency of neural networks can be drastically enhanced if the neuron outputs and/or the weights can be trained to take binary values ± 1 only, for which the MAC can be replaced by the simple XNOR operations. In this paper, we demonstrate an adiabatic training method that can binarize the fully-connected neural networks and the convolutional neural networks without modifying the network structure and size. This adiabatic training method only requires very minimal changes in training algorithms, and is tested in the following four tasks: the recognition of hand-writing numbers using a usual fully-connected network, the cat-dog recognition and the audio recognition using convolutional neural networks, the image recognition with 10 classes (CIFAR-10) using ResNet-20 and VGG-Small networks. In all tasks, the performance of the binary neural networks trained by the adiabatic method are almost identical to the networks trained using the conventional ReLU or Sigmoid activations with real-valued activations and weights. This adiabatic method can be easily applied to binarize different types of networks, and will increase the computational efficiency considerably and greatly simplify the deployment of neural networks. [ABSTRACT FROM AUTHOR]

Coronavirus 2019 (COVID-19) is a new acute respiratory disease that has spread rapidly throughout the world. In this paper, a lightweight convolutional neural network (CNN) model named multi-scale gated multi-head attention depthwise separable CNN (MGMADS-CNN) is proposed, which is based on attention mechanism and depthwise separable convolution. A multi-scale gated multi-head attention mechanism is designed to extract effective feature information from the COVID-19 X-ray and CT images for classification. Moreover, the depthwise separable convolution layers are adopted as MGMADS-CNN's backbone to reduce the model size and parameters. The LeNet-5, AlexNet, GoogLeNet, ResNet, VGGNet-16, and three MGMADS-CNN models are trained, validated and tested with tenfold cross-validation on X-ray and CT images. The results show that MGMADS-CNN with three attention layers (MGMADS-3) has achieved accuracy of 96.75% on X-ray images and 98.25% on CT images. The specificity and sensitivity are 98.06% and 96.6% on X-ray images, and 98.17% and 98.05% on CT images. The size of MGMADS-3 model is only 43.6 M bytes. In addition, the detection speed of MGMADS-3 on X-ray images and CT images are 6.09 ms and 4.23 ms for per image, respectively. It is proved that the MGMADS-3 can detect and classify COVID-19 faster with higher accuracy and efficiency. [ABSTRACT FROM AUTHOR]

In this work a novel architecture, named pseudo-softmax, to compute an approximated form of the softmax function is presented. This architecture can be fruitfully used in the last layer of Neural Networks and Convolutional Neural Networks for classification tasks, and in Reinforcement Learning hardware accelerators to compute the Boltzmann action-selection policy. The proposed pseudo-softmax design, intended for efficient hardware implementation, exploits the typical integer quantization of hardware-based Neural Networks obtaining an accurate approximation of the result. In the paper, a detailed description of the architecture is given and an extensive analysis of the approximation error is performed by using both custom stimuli and real-world Convolutional Neural Networks inputs. The implementation results, based on CMOS standard-cell technology, compared to state-of-the-art architectures show reduced approximation errors. [ABSTRACT FROM AUTHOR]