Search

In image classification with Deep Convolutional Neural Networks (DCNNs), the number of parameters in pointwise convolutions rapidly grows due to the multiplication of the number of filters by the number of input channels that come from the previous layer. Existing studies demonstrated that a subnetwork can replace pointwise convolutional layers with significantly fewer parameters and fewer floating-point computations, while maintaining the learning capacity. In this paper, we propose an improved scheme for reducing the complexity of pointwise convolutions in DCNNs for image classification based on interleaved grouped filters without divisibility constraints. The proposed scheme utilizes grouped pointwise convolutions, in which each group processes a fraction of the input channels. It requires a number of channels per group as a hyperparameter C h . The subnetwork of the proposed scheme contains two consecutive convolutional layers K and L, connected by an interleaving layer in the middle, and summed at the end. The number of groups of filters and filters per group for layers K and L is determined by exact divisions of the original number of input channels and filters by C h . If the divisions were not exact, the original layer could not be substituted. In this paper, we refine the previous algorithm so that input channels are replicated and groups can have different numbers of filters to cope with non exact divisibility situations. Thus, the proposed scheme further reduces the number of floating-point computations (11%) and trainable parameters (10%) achieved by the previous method. We tested our optimization on an EfficientNet-B0 as a baseline architecture and made classification tests on the CIFAR-10, Colorectal Cancer Histology, and Malaria datasets. For each dataset, our optimization achieves a saving of 76%, 89%, and 91% of the number of trainable parameters of EfficientNet-B0, while keeping its test classification accuracy. [ABSTRACT FROM AUTHOR]

With the development of Internet technology, short texts have gradually become the main medium for people to obtain information and communicate. Short text reduces the threshold of information production and reading by virtue of its short length, which is in line with the trend of fragmented reading in the context of the current fast-paced life. In addition, short texts contain emojis to make the communication immersive. However, short-text content means it contains relatively little information, which is not conducive to the analysis of sentiment characteristics. Therefore, this paper proposes a sentiment classification method based on the blending of emoticons and short-text content. Emoticons and short-text content are transformed into vectors, and the corresponding word vector and emoticon vector are connected into a sentencing matrix in turn. The sentence matrix is input into a convolution neural network classification model for classification. The results indicate that, compared with existing methods, the proposed method improves the accuracy of analysis. [ABSTRACT FROM AUTHOR]

Medical image fusion (MIF) has received painstaking attention due to its diverse medical applications in response to accurately diagnosing clinical images. Numerous MIF methods have been proposed to date, but the fused image suffers from poor contrast, non-uniform illumination, noise presence, and improper fusion strategies, resulting in an inadequate sparse representation of significant features. This paper proposes the morphological preprocessing method to address the non-uniform illumination and noise by the bottom-hat–top-hat strategy. Then, grey-principal component analysis (grey-PCA) is used to transform RGB images into gray images that can preserve detailed features. After that, the local shift-invariant shearlet transform (LSIST) method decomposes the images into the low-pass (LP) and high-pass (HP) sub-bands, efficiently restoring all significant characteristics in various scales and directions. The HP sub-bands are fed to two branches of the Siamese convolutional neural network (CNN) by process of feature detection, initial segmentation, and consistency verification to effectively capture smooth edges, and textures. While the LP sub-bands are fused by employing local energy fusion using the averaging and selection mode to restore the energy information. The proposed method is validated by subjective and objective quality assessments. The subjective evaluation is conducted by a user case study in which twelve field specialists verified the superiority of the proposed method based on precise details, image contrast, noise in the fused image, and no loss of information. The supremacy of the proposed method is further justified by obtaining 0.6836 to 0.8794, 0.5234 to 0.6710, and 3.8501 to 8.7937 gain for Q F AB , CRR, and AG and noise reduction from 0.3397 to 0.1209 over other methods for objective parameters. [ABSTRACT FROM AUTHOR]

Sentiment analysis (SA) is an important task in natural language processing in which convolutional neural networks (CNNs) have been successfully applied. However, most existing CNNs can only extract predefined, fixed-scale sentiment features and cannot synthesize flexible, multi-scale sentiment features. Moreover, these models' convolutional and pooling layers gradually lose local detailed information. In this study, a new CNN model based on residual network technology and attention mechanisms is proposed. This model exploits more abundant multi-scale sentiment features and addresses the loss of locally detailed information to enhance the accuracy of sentiment classification. It is primarily composed of a position-wise gated Res2Net (PG-Res2Net) module and a selective fusing module. The PG-Res2Net module can adaptively learn multi-scale sentiment features over a large range using multi-way convolution, residual-like connections, and position-wise gates. The selective fusing module is developed to fully reuse and selectively fuse these features for prediction. The proposed model was evaluated using five baseline datasets. The experimental results demonstrate that the proposed model surpassed the other models in performance. In the best case, the model outperforms the other models by up to 1.2%. Ablation studies and visualizations further revealed the model's ability to extract and fuse multi-scale sentiment features. [ABSTRACT FROM AUTHOR]

Decoding motor imagery (MI) electroencephalogram (EEG) signals for brain-computer interfaces (BCIs) is a challenging task because of the severe non-stationarity of perceptual decision processes. Recently, deep learning techniques have had great success in EEG decoding because of their prominent ability to learn features from raw EEG signals automatically. However, the challenge that the deep learning method faces is that the shortage of labeled EEG signals and EEGs sampled from other subjects cannot be used directly to train a convolutional neural network (ConvNet) for a target subject. To solve this problem, in this paper, we present a novel conditional domain adaptation neural network (CDAN) framework for MI EEG signal decoding. Specifically, in the CDAN, a densely connected ConvNet is firstly applied to obtain high-level discriminative features from raw EEG time series. Then, a novel conditional domain discriminator is introduced to work as an adversarial with the label classifier to learn commonly shared intra-subjects EEG features. As a result, the CDAN model trained with sufficient EEG signals from other subjects can be used to classify the signals from the target subject efficiently. Competitive experimental results on a public EEG dataset (High Gamma Dataset) against the state-of-the-art methods demonstrate the efficacy of the proposed framework in recognizing MI EEG signals, indicating its effectiveness in automatic perceptual decision decoding. [ABSTRACT FROM AUTHOR]

Recently, convolutional neural network (CNN) based on the encoder-decoder structure have been successfully applied to image dehazing. However, these CNN based dehazing methods have two limitations: First, these dehazing models are large in size with enormous parameters, which not only consumes much GPU memory, but also is hard to train from scratch. Second, these models, which ignore the structural information at different resolutions of intermediate layers, cannot capture informative texture and edge information for dehazing by stacking more layers. In this paper, we propose a light-weight end-to-end network named the residual dense pyramid network (RDPN) to address the above problems. To exploit the structural information at different resolutions of intermediate layers fully, a new residual dense pyramid (RDP) is proposed as a building block. By introducing a dense information fusion layer and the residual learning module, the RDP can maximize the information flow and extract local features. Furthermore, the RDP further learns the structural information from intermediate layers via a multiscale pyramid fusion mechanism. To reduce the number of network parameters and to ease the training process, we use one RDP in the encoder and two RDPs in the decoder, following a multilevel pyramid pooling layer for incorporating global context features before estimating the final result. The extensive experimental results on a synthetic dataset and real-world images demonstrate that the new RDPN achieves favourable performance compared with some state-of-the-art methods, e.g., the recent densely connected pyramid dehazing network, the all-in-one dehazing network, the enhanced pix2pix dehazing network, pixel-based alpha blending, artificial multi-exposure image fusions and the genetic programming estimator, in terms of accuracy, run time and number of parameters. To be specific, RDPN outperforms all of the above methods in terms of PSNR by at least 4.25 dB. The run time of the proposed method is 0.021 s, and the number of parameters is 1,534,799, only 6% of that used by the densely connected pyramid dehazing network. [ABSTRACT FROM AUTHOR]

Image retrieval based on a convolutional neural network (CNN) has attracted great attention among researchers because of the high performance. The pooling method has become a research hotpot in the task of image retrieval in recent years. In this paper, we propose the feature distribution entropy (FDE) to measure the difference of regional distribution information in the feature maps from CNNs. We propose a novel pooling method, which fuses our proposed FDE with region maximum activations of convolutions (R-MAC) features to improve the performance of image retrieval, as it takes the advantage of regional distribution information in the feature maps. Compared with the descriptors computed by R-MAC pooling, our proposed method considers not only the most significant feature values of each region in feature map, but also the distribution difference in different regions. We utilize the histogram of feature values to calculate regional distribution entropy and concatenate the regional distribution entropy into FDE, which is further normalized and fused with R-MAC feature vectors by weighted summation to generate the final feature descriptors. We have conducted experiments on public datasets and the results demonstrate that our proposed method could produce better retrieval performances than existing state-of-the-art algorithms. Further, higher performance could be achieved by performing these post-processing on the improved feature descriptors. [ABSTRACT FROM AUTHOR]

Detecting and classifying ships based on radiated noise provide practical guidelines for the reduction of underwater noise footprint of shipping. In this paper, the detection and classification are implemented by auditory inspired convolutional neural networks trained from raw underwater acoustic signal. The proposed model includes three parts. The first part is performed by a multi-scale 1D time convolutional layer initialized by auditory filter banks. Signals are decomposed into frequency components by convolution operation. In the second part, the decomposed signals are converted into frequency domain by permute layer and energy pooling layer to form frequency distribution in auditory cortex. Then, 2D frequency convolutional layers are applied to discover spectro-temporal patterns, as well as preserve locality and reduce spectral variations in ship noise. In the third part, the whole model is optimized with an objective function of classification to obtain appropriate auditory filters and feature representations that are correlative with ship categories. The optimization reflects the plasticity of auditory system. Experiments on five ship types and background noise show that the proposed approach achieved an overall classification accuracy of 79.2%, which improved by 6% compared to conventional approaches. Auditory filter banks were adaptive in shape to improve accuracy of classification. [ABSTRACT FROM AUTHOR]

Deep convolutional neural networks (ConvNets), which are at the heart of many new emerging applications, achieve remarkable performance in audio and visual recognition tasks. Unfortunately, achieving accuracy often implies significant computational costs, limiting deployability. In modern ConvNets it is typical for the convolution layers to consume the vast majority of computational resources during inference. This has made the acceleration of these layers an important research area in academia and industry. In this paper, we examine the effects of co-optimizing the internal structures of the convolutional layers and underlying implementation of fundamental convolution operation. We demonstrate that a combination of these methods can have a big impact on the overall speedup of a ConvNet, achieving a ten-fold increase over baseline. We also introduce a new class of fast one-dimensional (1D) convolutions for ConvNets using the Toom–Cook algorithm. We show that our proposed scheme is mathematically well-grounded, robust, and does not require any time-consuming retraining, while still achieving speedups solely from convolutional layers with no loss in baseline accuracy. [ABSTRACT FROM AUTHOR]

The differential diagnosis of epileptic seizures (ES) and psychogenic non-epileptic seizures (PNES) may be difficult, due to the lack of distinctive clinical features. The interictal electroencephalographic (EEG) signal may also be normal in patients with ES. Innovative diagnostic tools that exploit non-linear EEG analysis and deep learning (DL) could provide important support to physicians for clinical diagnosis. In this work, 18 patients with new-onset ES (12 males, 6 females) and 18 patients with video-recorded PNES (2 males, 16 females) with normal interictal EEG at visual inspection were enrolled. None of them was taking psychotropic drugs. A convolutional neural network (CNN) scheme using DL classification was designed to classify the two categories of subjects (ES vs. PNES). The proposed architecture performs an EEG time-frequency transformation and a classification step with a CNN. The CNN was able to classify the EEG recordings of subjects with ES vs. subjects with PNES with 94.4% accuracy. CNN provided high performance in the assigned binary classification when compared to standard learning algorithms (multi-layer perceptron, support vector machine, linear discriminant analysis and quadratic discriminant analysis). In order to interpret how the CNN achieved this performance, information theoretical analysis was carried out. Specifically, the permutation entropy (PE) of the feature maps was evaluated and compared in the two classes. The achieved results, although preliminary, encourage the use of these innovative techniques to support neurologists in early diagnoses. [ABSTRACT FROM AUTHOR]

The image quality evaluation method, based on the convolutional neural network (CNN), achieved good evaluation performance. However, this method can easily lead the visual quality of image sub-blocks to change with the spatial position after the image is processed by various distortions. Consequently, the visual quality of the entire image is difficult to reflect objectively. On this basis, this study combines wavelet transform and CNN method to propose an image quality evaluation method based on wavelet CNN. The low-frequency, horizontal, vertical, and diagonal sub-band images decomposed by wavelet transform are selected as the inputs of convolution neural network. The feature information in multiple directions is extracted by convolution neural network. Then, the information entropy of each sub-band image is calculated and used as the weight of each sub-band image quality. Finally, the quality evaluation values of four sub-band images are weighted and fused to obtain the visual quality values of the entire image. Experimental results show that the proposed method gains advantage from the global and local information of the image, thereby further improving its effectiveness and generalization. [ABSTRACT FROM AUTHOR]

The automatic detection of pulse during out-of-hospital cardiac arrest (OHCA) is necessary for the early recognition of the arrest and the detection of return of spontaneous circulation (end of the arrest). The only signal available in every single defibrillator and valid for the detection of pulse is the electrocardiogram (ECG). In this study we propose two deep neural network (DNN) architectures to detect pulse using short ECG segments (5 s), i.e., to classify the rhythm into pulseless electrical activity (PEA) or pulse-generating rhythm (PR). A total of 3914 5-s ECG segments, 2372 PR and 1542 PEA, were extracted from 279 OHCA episodes. Data were partitioned patient-wise into training (80%) and test (20%) sets. The first DNN architecture was a fully convolutional neural network, and the second architecture added a recurrent layer to learn temporal dependencies. Both DNN architectures were tuned using Bayesian optimization, and the results for the test set were compared to state-of-the art PR/PEA discrimination algorithms based on machine learning and hand crafted features. The PR/PEA classifiers were evaluated in terms of sensitivity (Se) for PR, specificity (Sp) for PEA, and the balanced accuracy (BAC), the average of Se and Sp. The Se/Sp/BAC of the DNN architectures were 94.1%/92.9%/93.5% for the first one, and 95.5%/91.6%/93.5% for the second one. Both architectures improved the performance of state of the art methods by more than 1.5 points in BAC. [ABSTRACT FROM AUTHOR]

Drug-drug interactions (DDIs) may bring huge health risks and dangerous effects to a patient's body when taking two or more drugs at the same time or within a certain period of time. Therefore, the automatic extraction of unknown DDIs has great potential for the development of pharmaceutical agents and the safety of drug use. In this article, we propose a novel recurrent hybrid convolutional neural network (RHCNN) for DDI extraction from biomedical literature. In the embedding layer, the texts mentioning two entities are represented as a sequence of semantic embeddings and position embeddings. In particular, the complete semantic embedding is obtained by the information fusion between a word embedding and its contextual information which is learnt by recurrent structure. After that, the hybrid convolutional neural network is employed to learn the sentence-level features which consist of the local context features from consecutive words and the dependency features between separated words for DDI extraction. Lastly but most significantly, in order to make up for the defects of the traditional cross-entropy loss function when dealing with class imbalanced data, we apply an improved focal loss function to mitigate against this problem when using the DDIExtraction 2013 dataset. In our experiments, we achieve DDI automatic extraction with a micro F-score of 75.48% on the DDIExtraction 2013 dataset, outperforming the state-of-the-art approach by 2.49%. [ABSTRACT FROM AUTHOR]

The processing and analyzing of multimedia data has become a popular research topic due to the evolution of deep learning. Deep learning has played an important role in addressing many challenging problems, such as computer vision, image recognition, and image detection, which can be useful in many real-world applications. In this study, we analyzed visual features of images to detect advertising images from scanned images of various magazines. The aim is to identify key features of advertising images and to apply them to real-world application. The proposed work will eventually help improve marketing strategies, which requires the classification of advertising images from magazines. We employed convolutional neural networks to classify scanned images as either advertisements or non-advertisements (i.e., articles). The results show that the proposed approach outperforms other classifiers and the related work in terms of accuracy. [ABSTRACT FROM AUTHOR]