Typical Convolutional Neural Network Research Articles

Encrypted Network Traffic Classification Using Deep and Parallel Network-In-Network Models

Network traffic classification analyzes received data packets to identify distinct application or traffic kinds. This research describes a neural network model that uses deep and parallel network-in-network (NIN) architectures to classify encrypted network data. In comparison to typical convolutional neural networks (CNN), NIN uses a micro network after each convolution layer to improve local modeling. Furthermore, NIN uses global average pooling instead of traditional fully connected layers before final classification, resulting in a considerable reduction in the number of model parameters. Our suggested solution uses deep NIN models with several MLP convolutional layers to map fixed-length packet vectors to application or traffic labels. Furthermore, a parallel decision method is created to build two sub-networks to process packet headers and packet bodies separately, taking into account that they may include different types of evidence for classification. Our investigations on the ''ISCX VPN-nonVPN'' encrypted traffic dataset demonstrate that NIN models can achieve a better balance between classification accuracy and model complexity than standard CNNs. The parallel decision technique can increase the accuracy of a single NIN model for classifying encrypted network data. Finally, the test set F1 scores of 0.983 and 0.985 are obtained for traffic characterisation and application identification, respectively

Pure Vision Transformer (CT-ViT) with Noise2Neighbors Interpolation for Low-Dose CT Image Denoising.

Convolutional neural networks (CNN) have been used for a wide variety of deep learning applications, especially in computer vision. For medical image processing, researchers have identified certain challenges associated with CNNs. These challenges encompass the generation of less informative features, limitations in capturing both high and low-frequency information within feature maps, and the computational cost incurred when enhancing receptive fields by deepening the network. Transformers have emerged as an approach aiming to address and overcome these specific limitations of CNNs in the context of medical image analysis. Preservation of all spatial details of medical images is necessary to ensure accurate patient diagnosis. Hence, this research introduced the use of a pure Vision Transformer (ViT) for a denoising artificial neural network for medical image processing specifically for low-dose computed tomography (LDCT) image denoising. The proposed model follows a U-Net framework that contains ViT modules with the integration of Noise2Neighbor (N2N) interpolation operation. Five different datasets containing LDCT and normal-dose CT (NDCT) image pairs were used to carry out this experiment. To test the efficacy of the proposed model, this experiment includes comparisons between the quantitative and visual results among CNN-based (BM3D, RED-CNN, DRL-E-MP), hybrid CNN-ViT-based (TED-Net), and the proposed pure ViT-based denoising model. The findings of this study showed that there is about 15-20% increase in SSIM and PSNR when using self-attention transformers than using the typical pure CNN. Visual results also showed improvements especially when it comes to showing fine structural details of CT images.

Deep Learning Estimating of Epicentral Distance for Earthquake Early Warning Systems

ABSTRACT To enhance the performance of earthquake early warning (EEW) systems that aim to issue alerts as quickly as possible, it is crucial to improve the accuracy of the epicentral distance Δ estimated via the single-station method. Although the conventional method estimates Δ from the slope of the initial P-wave envelope, this study applies deep learning techniques that can extract a variety of information from the waveform data. By analyzing ∼20,000 records observed at Kyoshin Network stations in Japan, the convolutional neural network (CNN) method achieved higher accuracy than the conventional method. Increasing the data length or the number of iterations of convolution, activation, and pooling layers in the typical CNN model did not significantly improve the accuracy of Δ estimation. An automatic structure search (AutoSS) technique, in which model structure and hyperparameters are randomly varied, was employed to identify models that yield higher accuracy. A typical CNN model was used as the initial structure. The models obtained through this technique showed improved accuracy with increased data length or computational cost. The models that delivered the highest accuracy among those generated using the AutoSS technique outperformed the typical CNN model in terms of accuracy, although their computational costs were comparable. The AutoSS technique offers a significant advantage in that it allows the selection of a model that matches the computational capabilities of the hardware used for its implementation, thereby ensuring optimal computational efficiency. This exhibits that deep learning technologies can be used to improve the performance of EEW systems.

Brain tumor prediction and analysis based on simple CNN

The field of brain tumor prediction aims to detect potentially malignant tumors before overt symptoms appear. Early tumor detection can help tailor treatment pathways. To this end, this paper introduces a convolutional neural network (CNN) to construct a simple CNN analysis model. Subtle architectural changes to typical CNN designs are aimed at navigating and solving complex problems unique to medical imaging. Specifically, first, a preprocessing stage converts raw Magnetic Resonance Imaging (MRI) scans into analyzable data. This enables the model to delve into complex patterns in each image. Second, build a simple CNN model and evaluate it. Additionally, comparisons are made with established benchmarks set by traditional CNN architectures. The study further utilizes many available annotated datasets to develop a data augmentation paradigm that improves model resilience and generalizability. At the same time, the study deploys rigorous protocols to address and contain the ubiquitous overfitting threat, thereby enhancing the general applicability of the solution across a variety of healthcare settings. Empirical evaluations shed light on the significant superiority of the proposed simple CNN, especially in specific imaging environments and data structures. At the heart of this research is not only to expand technological horizons but also to reimagine clinical diagnostics to improve the standard of patient care.

ADFCNN: Attention-Based Dual-Scale Fusion Convolutional Neural Network for Motor Imagery Brain-Computer Interface.

Convolutional neural networks (CNNs) have been successfully applied to motor imagery (MI)-based brain-computer interface (BCI). Nevertheless, single-scale CNN fail to extract abundant information over a wide spectrum from EEG signals, while typical multi-scale CNNs cannot effectively fuse information from different scales with concatenation-based methods. To overcome these challenges, we propose a new scheme equipped with attention-based dual-scale fusion convolutional neural network (ADFCNN), which jointly extracts and fuses EEG spectral and spatial information at different scales. This scheme also provides novel insight through self-attention for effective information fusion from different scales. Specifically, temporal convolutions with two different kernel sizes identify EEG μ and β rhythms, while spatial convolutions at two different scales generate global and detailed spatial information, respectively, and the self-attention mechanism performs feature fusion based on the internal similarity of the concatenated features extracted by the dual-scale CNN. The proposed scheme achieves the superior performance compared with state-of-the-art methods in subject-specific motor imagery recognition on BCI Competition IV dataset 2a, 2b and OpenBMI dataset, with the cross-session average classification accuracies of 79.39% and significant improvements of 9.14% on BCI-IV2a, 87.81% and 7.66% on BCI-IV2b, 65.26% and 7.2% on OpenBMI dataset, and the within-session average classification accuracies of 86.87% and significant improvements of 10.89% on BCI-IV2a, 87.26% and 8.07% on BCI-IV2b, 84.29% and 5.17% on OpenBMI dataset, respectively. What is more, ablation experiments are conducted to investigate the mechanism and demonstrate the effectiveness of the dual-scale joint temporal-spatial CNN and self-attention modules. Visualization is also used to reveal the learning process and feature distribution of the model.

Deep understanding of radiology reports: leveraging dynamic convolution in chest X-ray images

PurposeThe purpose of this study is to investigate and demonstrate the advancements achieved in the field of chest X-ray image captioning through the utilization of dynamic convolutional encoder–decoder networks (DyCNN). Typical convolutional neural networks (CNNs) are unable to capture both local and global contextual information effectively and apply a uniform operation to all pixels in an image. To address this, we propose an innovative approach that integrates a dynamic convolution operation at the encoder stage, improving image encoding quality and disease detection. In addition, a decoder based on the gated recurrent unit (GRU) is used for language modeling, and an attention network is incorporated to enhance consistency. This novel combination allows for improved feature extraction, mimicking the expertise of radiologists by selectively focusing on important areas and producing coherent captions with valuable clinical information.Design/methodology/approachIn this study, we have presented a new report generation approach that utilizes dynamic convolution applied Resnet-101 (DyCNN) as an encoder (Verelst and Tuytelaars, 2019) and GRU as a decoder (Dey and Salemt, 2017; Pan et al., 2020), along with an attention network (see Figure 1). This integration innovatively extends the capabilities of image encoding and sequential caption generation, representing a shift from conventional CNN architectures. With its ability to dynamically adapt receptive fields, the DyCNN excels at capturing features of varying scales within the CXR images. This dynamic adaptability significantly enhances the granularity of feature extraction, enabling precise representation of localized abnormalities and structural intricacies. By incorporating this flexibility into the encoding process, our model can distil meaningful and contextually rich features from the radiographic data. While the attention mechanism enables the model to selectively focus on different regions of the image during caption generation. The attention mechanism enhances the report generation process by allowing the model to assign different importance weights to different regions of the image, mimicking human perception. In parallel, the GRU-based decoder adds a critical dimension to the process by ensuring a smooth, sequential generation of captions.FindingsThe findings of this study highlight the significant advancements achieved in chest X-ray image captioning through the utilization of dynamic convolutional encoder–decoder networks (DyCNN). Experiments conducted using the IU-Chest X-ray datasets showed that the proposed model outperformed other state-of-the-art approaches. The model achieved notable scores, including a BLEU_1 score of 0.591, a BLEU_2 score of 0.347, a BLEU_3 score of 0.277 and a BLEU_4 score of 0.155. These results highlight the efficiency and efficacy of the model in producing precise radiology reports, enhancing image interpretation and clinical decision-making.Originality/valueThis work is the first of its kind, which employs DyCNN as an encoder to extract features from CXR images. In addition, GRU as the decoder for language modeling was utilized and the attention mechanisms into the model architecture were incorporated.

Deep Learning Based on Enhanced MRI T1 Imaging to Differentiate Small-cell and Non-small-cell Primary Lung Cancers in Patients with Brain Metastases.

To differentiate the primary small-cell lung cancer (SCLC) and non-small-cell lung cancer (NSCLC) for patients with brain metastases (BMs) based on a deep learning (DL) model using contrast-enhanced magnetic resonance imaging (MRI) T1 weighted (T1CE) images. Out of 711 patients with BMs of lung cancer origin (SCLC 232, NSCLC 479), the MRI datasets of 192 patients (lesions' widths and heights > 30 pixels) with BMs from lung cancer (73 SCLC and 119 NSCLC) confirmed pathologically were enrolled, retrospectively. A typical convolutional neural network ResNet18 was applied for the automatic classification of BMs lesions from lung cancer based on T1CE images, with training and testing groups randomized per patient to eliminate learning bias. A 5-fold cross-validation was performed to evaluate the classification of the model. The receiver operating characteristic (ROC) curve, accuracy, precision, recall and f1 score were calculated. For a 5-fold cross-validation test, the DL model achieved AUCs of 0.8019 and 0.8024 for SCLC and NSCLC patients with BMs, respectively, and a mean overall accuracy of 0.7515±0.04. The DL model performed well in differentiating the primary SCLC and NSCLC with BMs. The proposed DL model is feasible and effective in differentiating the pathological subtypes of SCLC and NSCLC causing BMs, which may be used as a new tool for oncologists to diagnose noninvasively BMs and guide therapy based on the imaging structure of tumors.

Deep learning-based prediction of the remaining time and future distribution of pebble flow from real-scene images

Pebble flow dynamics is a crucial issue for designing and operating pebble bed reactors. The existing experimental or simulation methods are often associated with high time, resource, and effort costs. Therefore, image-based deep learning methods are explored to make real-time predictions of pebble flow dynamics directly from images. Real-scene images, captured by the high-speed camera during pebble flow experiments are used as the dataset. This paper proposes an RT-Net model based on Convolutional Neural Network (CNN) to predict the remaining time of pebble flow from experimental images. The core of the RT-Net model is the mergeable multi-branch convolutional component called ConvBlock, which effectively improves accuracy and reduces computational costs compared to traditional convolutional operators. Results show that this model is superior in both accuracy and efficiency metrics compared with typical convolutional neural networks (like AlexNet and VGGs). The proportion of test sets with prediction error within 0.05 s reaches 96.7 %, and the parameters count and inference time are 5.02 M and 1.99 ms respectively. Furthermore, to anticipate the pebble distribution at a given future time, a dual-input PreNet that combines CNN and Generative Adversarial Network (GAN) is designed, where the input includes the current pebble flow image and an arbitrarily chosen temporal displacement Δt. Targeted evaluation metrics, such as Target Similarity (TS) and Equivalent Target Similarity (TSeq) are proposed to quantitatively evaluate the model’s performance. Results indicate that the Pre-Net model can make quite satisfactory predictions of the future distribution of most data, while more efforts such as a task-specific loss function are encouraged to achieve better performance.

Classification of Epileptic Seizure Types Using Multiscale Convolutional Neural Network and Long Short-Term Memory

The accurate classification of seizure types using electroencephalography (EEG) signals plays a vital role in determining a precise treatment plan and therapy for epilepsy patients. Among the available deep network models, Convolutional Neural Networks (CNNs) are the most widely adopted models for learning and representing EEG signals. However, typical CNNs have high computational complexity, leading to overfitting problems. This paper proposes the design of two effective, lightweight deep network models; the 1D multiscale neural network (1D-MSCNet) model and the Long Short-term Memory (LSTM)-based compact CNN (EEG-LSTMNet) model. The 1D-MSCNet model comprises three modules: a spectral–temporal convolution module, a spatial convolution module, and a classification module. It extracts features from input EEG trials at multiple frequency/time ranges, identifying relationships between the spatial distribution of their channels. The EEG-LSTMNet model includes three convolutional layers, namely temporal, depthwise, and separable layers, a single LSTM layer, and two fully connected classification layers to extract discriminative EEG feature representations. Both models have been applied to the same EEG trials collected from the Temple University Hospital (TUH) database. Results revealed F1-score values of 96.9% and 98.4% for the 1D-MSCNet and EEG-LSTMNet, respectively. Based on the demonstrated outcomes, both models outperform related state-of-the-art methods due to their architectures’ adoption of 1D modules and layers that reduce the computational effort needed, solve the overfitting problem, and enhance classification efficiency. Hence, both models could be valuable additions for neurologists to help them decide upon precise treatments and drugs for patients depending on their type of seizure.

Retrieving Atmospheric Gas Profiles Using FY-3E/HIRAS-II Infrared Hyperspectral Data by Neural Network Approach

The observed radiation data from the second-generation Hyperspectral Infrared Atmospheric Sounder (HIRAS-II) on the Fengyun-3E (FY-3E) satellite contain useful vertical atmosphere information which can distinguish and retrieve vertical profiles of atmospheric gas components including ozone (O3), carbon monoxide (CO), and methane (CH4). This paper utilizes FY-3E/HIRAS-II observational data to optimize each gas channel using the improved Optimal Sensitivity Profile method (OSP) channel algorithm and establishes a typical convolutional neural network model (CNN) and a representative U-shaped network model (UNET) with deep features and shallow feature links to perform atmospheric profile retrieval calculations of O3, CO, and CH4. We chose the clear sky data of the Indian and its southern seas in December 2021 and January 2022, with reanalysis data from European Center for Medium-Range Weather Forecasts Reanalysis v5 (ERA5) and European Center for Medium-Range Weather Forecasts Atmospheric Composition Reanalysis v4 (EAC4) serving as the reference values. The retrieval outcomes were then compared against advanced numerical forecast models including the Whole Atmosphere Community Climate Model (WACCM), Global Forecast System (GFS), and satellite products from an Atmospheric Infrared Sounder (AIRS) and Infrared Atmospheric Sounding Interferometer (IASI). Experimental results show that the generalization ability and retrieval accuracy of CNN are slightly higher compared with UNET. For O3 profile retrieval, the mean percentage error (MPE) of the whole layers for CNN and UNET data in relation to ERA5 data was less than 8%, while the root-mean-square error (RMSE) was below 1.5 × 10−7 kg/kg; for CH4 profile retrieval, the MPE of the whole layers for CNN and UNET data in relation to EAC4 data was less than 0.7%, while the RMSE was below 1.5 × 10−8 kg/kg. The retrieval of O3 and CH4 are resulted in a significant improvement compared to the forecast data and satellite products in most pressure levels; for CO profile retrieval, the MPE of the whole layers for CNN and UNET data in relation to EAC4 data was less than 11%, while the RMSE was below 4 × 10−8 kg/kg. The error of the CO retrieval results was higher than that of the forecast data at the pressure level of 200~500 hPa and lower than that of similar satellite products with most pressure levels. The experiments indicated that the neural network method effectively determines the atmospheric gas profiles using infrared hyperspectral data, exhibiting a positive performance in accuracy and retrieval speed.

Predicting micromechanical properties of cement paste from backscattered electron (BSE) images by computer vision

This paper employs computer vision techniques to predict the micromechanical properties (i.e., elastic modulus and hardness) of cement paste based on an input of Backscattered Electron (BSE) images. A dataset comprising 40,000 nanoindentation tests and 40,000 BSE micrographs was built by express nanoindentation test and Scanning Electron Microscopy (SEM). A Residual Convolutional Neural Network (Res-Net) model, which differs from a typical Convolutional Neural Network (CNN) architecture by a shortcut connection, was employed and compared with a simple table model. The models were trained, tuned, and tested over a training, validation and testing set comprising 70%, 15% and 15% of the 40,000 data pairs, respectively. The following conclusions were drawn: 1) Express nanoindentation tests can provide reliable information for cement paste. Deconvolution based on Gaussian Mixture Model (GMM) can obtain almost invariant statistics for each phase; 2) Based on averaged greyscale values of each BSE image, a table model can predict the elastic modulus and hardness with R2 of 0.80 and 0.83, respectively; 3) Based on the intensity of each pixel as well as their patterns in each BSE image, the Res-Net model can predict the elastic modulus and hardness with a R2 of 0.85 and 0.88, respectively. Deconvolution of the Res-Net prediction obtains similar invariant statistics as derived by the nanoindentation tests, which gives strong evidence of the applicability of the Res-Net model.

Masked Visual Transformer for Efficient Training with Small Dataset

Vision Transformers (ViTs) are becoming an architectural paradigm replacing the convolutional neural networks (CNNs) in computer vision. ViTs offer competitive performances with respect to CNNs, but they, especially the vanilla ViTs, are hungrier for data than the typical CNNs as the short of the common inductive bias of convolution. Recently, few works focused on training vanilla ViTs efficiently with small datasets. In this paper, we perform research on training vanilla ViTs with small dataset containing thousands of images, and propose a method that is applied to self-supervised pretraining stage. The proposed method combines parametric instance discrimination with CutMix and Multi-crop. Furthermore, we introduce image masking to reduce the overfitting of pretraining on small dataset. State-of-the-art results are achieved by our method for training from scratch based on vanilla ViT backbones on seven small-scale datasets. The transferring performance of our method is also tested on small datasets, and results show that it is improved significantly.

Energy efficiency and savings analysis with multirate sampling for petrochemical process using convolutional neural network-based transfer learning

Energy efficiency in the petrochemical industry is crucial in reducing energy consumption and environmental impact. An accurate energy efficiency model will provide valuable insight for supporting operational adjustment decisions. In practice, due to inconsistent sampling intervals in the petrochemical industry, the traditional approach for obtaining energy efficiency may be unreliable and difficult to handle these multirate data characteristics. Therefore, in this paper, a multi-channel convolutional neural network model integrating a model parameter-based transfer learning approach is proposed to improve the prediction of energy efficiency under inconsistent sampling intervals. The multi-channel structure aims to recognize a different pattern from the dataset by convolving the information along the time dimension. Concurrently, transfer learning allows the model to learn a new pattern of input after the model is fully trained. Finally, the performance for energy efficiency prediction and saving analysis is validated by applying it to the vinyl chloride monomer production case study. The result shows that the proposed model outperformed traditional models and typical convolutional neural network structures in terms of accuracy and reproducibility, with an r-square of 0.97. The utilization of transfer learning prevents a significant drop in performance and enhances adaptability in model learning on real-time energy tracking. Moreover, the energy gap analysis of the prediction result identified a significant energy-saving potential, which would decrease annual energy consumption by 7.25% on average and a 5,709-ton reduction in carbon dioxide emissions.

Machine learning-based prediction for single-cell mechanics

Single-cell mechanics have gained much attention due to its importance in a broad range of biological applications. Different experimental approaches have been used for measuring the mechanical properties of individual cells. However, the technical demands and time-consuming nature of these procedures have limited the throughput of single-cell measurement, necessitating the development of alternative computational approaches. Recently, single-cell deformability can be predicted using a convolutional neural network (CNN) model, shedding a light on using machine learning (ML) algorithms for high-throughput characterizations of single-cell mechanical properties. In this work, we developed a novel ML-based computational framework that can reproduce a physical microfluidic system to investigate the individual cell's deformability. The datasets for the training and testing of our model were generated using high-fidelity fluid-structure interaction (FSI) simulations. Our FSI-based ML approach of adopting CNN algorithms demonstrated a highly accurate prediction for the membrane stiffness of a microcapsule (maximum R2 = 0.98) based on its deformed shape. In this paper, we show that by applying physical constraints including the microcapsule's total surface area and total volume, we were able to build a physics-constrained ML model that possesses better convergence and higher stability during both training and validation. Finally, we found that ML models that used the three-dimensional geometry of the capsule as input could outperform the typical CNN models that relied solely on the two-dimensional images. We expect that this physics-constrained computational framework will serve as a basis for developing future tools for real-time biological applications through the integration of high-fidelity simulations with ML algorithms.

The Development of Image Classification Algorithms Based on CNNs

Image classification is a method to process the given images by picking up the distinct features of different kinds of images to distinguish different targets in the image. The classification of images analyzes the given images quantitatively by using computers and divides the image or areas contained in the image into several categories instead of explaining them by human’s visualization. Image classification is an important research direction in computer vision technology and the basis for other applications such as image detection, behavior analysis, and object tracking. With the arrival of big data and the improvement of computer power, Deep Learning (DL) has swept the world, and the Convolutional Neural Network (CNN) image classification method has broken down the limitations of traditional image classification methods and has become a currently mainstream image classification algorithm. Thereinto, some typical architectures e.g., MobileNet, ResNet and VGG are attracted a lot of attention. This paper will review the development of the image classification and introduce some typical CNNs.

Tooth Type Enhanced Transformer for Children Caries Diagnosis on Dental Panoramic Radiographs

The objective of this study was to introduce a novel deep learning technique for more accurate children caries diagnosis on dental panoramic radiographs. Specifically, a swin transformer is introduced, which is compared with the state-of-the-art convolutional neural network (CNN) methods that are widely used for caries diagnosis. A tooth type enhanced swin transformer is further proposed by considering the differences among canine, molar and incisor. Modeling the above differences in swin transformer, the proposed method was expected to mine domain knowledge for more accurate caries diagnosis. To test the proposed method, a children panoramic radiograph database was built and labeled with a total of 6028 teeth. Swin transformer shows better diagnosis performance compared with typical CNN methods, which indicates the usefulness of this new technique for children caries diagnosis on panoramic radiographs. Furthermore, the proposed tooth type enhanced swin transformer outperforms the naive swin transformer with the accuracy, precision, recall, F1 and area-under-the-curve being 0.8557, 0.8832, 0.8317, 0.8567 and 0.9223, respectively. This indicates that the transformer model can be further improved with a consideration of domain knowledge instead of a copy of previous transformer models designed for natural images. Finally, we compare the proposed tooth type enhanced swin transformer with two attending doctors. The proposed method shows higher caries diagnosis accuracy for the first and second primary molars, which may assist dentists in caries diagnosis.

From crowd to herd counting: How to precisely detect and count African mammals using aerial imagery and deep learning?

Rapid growth of human populations in sub-Saharan Africa has led to a simultaneous increase in the number of livestock, often leading to conflicts of use with wildlife in protected areas. To minimize these conflicts, and to meet both communities’ and conservation goals, it is therefore essential to monitor livestock density and their land use. This is usually done by conducting aerial surveys during which aerial images are taken for later counting. Although this approach appears to reduce counting bias, the manual processing of images is time-consuming. The use of dense convolutional neural networks (CNNs) has emerged as a very promising avenue for processing such datasets. However, typical CNN architectures have detection limits for dense herds and close-by animals. To tackle this problem, this study introduces a new point-based CNN architecture, HerdNet, inspired by crowd counting. It was optimized on challenging oblique aerial images containing herds of camels (Camelus dromedarius), donkeys (Equus asinus), sheep (Ovis aries) and goats (Capra hircus), acquired over heterogeneous arid landscapes of the Ennedi reserve (Chad). This approach was compared to an anchor-based architecture, Faster-RCNN, and a density-based, adapted version of DLA-34 that is typically used in crowd counting. HerdNet achieved a global F1 score of 73.6 % on 24 megapixels images, with a root mean square error of 9.8 animals and at a processing speed of 3.6 s, outperforming the two baselines in terms of localization, counting and speed. It showed better proximity-invariant precision while maintaining equivalent recall to that of Faster-RCNN, thus demonstrating that it is the most suitable approach for detecting and counting large mammals at close range. The only limitation of HerdNet was the slightly weaker identification of species, with an average confusion rate approximately 4 % higher than that of Faster-RCNN. This study provides a new CNN architecture that could be used to develop an automatic livestock counting tool in aerial imagery. The reduced image analysis time could motivate more frequent flights, thus allowing a much finer monitoring of livestock and their land use.

Detection of Atrial Fibrillation from Variable-duration ECG Signal based on Time-adaptive Densely Network and Feature Enhancement Strategy.

Atrial fibrillation (AF) is one of the clinic's most common arrhythmias with high morbidity and mortality. Developing an intelligent auxiliary diagnostic model of AF based on a body surface electrocardiogram (ECG) is necessary. Convolutional neural network (CNN) is one of the most commonly used models for AF recognition. However, typical CNN is not compatible with variable-duration ECG, so it is hard to demonstrate its universality and generalization in practical applications. Hence, this paper proposes a novel Time-adaptive densely network named MP-DLNet-F. The MP-DLNet module solves the problem of incompatibility between variable-duration ECG and 1D-CNN. In addition, the feature enhancement module and data imbalance processing module are respectively used to enhance the perception of temporal-quality information and decrease the sensitivity to data imbalance. The experimental results indicate that the proposed MP-DLNet-F achieved 87.98% classification accuracy, and F1-score of 0.847 on the CinC2017 database for 10-second cropped/padded single-lead ECG fragments. Furthermore, we deploy transfer learning techniques to test heterogeneous datasets, and in the CPSC2018 12-lead dataset, the method improved the average accuracy and F1-score by 21.81% and 16.14%, respectively. Experimental results indicate that our method can update the constructed model's parameters and precisely forecast AF with different duration distributions and lead distributions. Combining these advantages, MP-DLNet-F can exemplify all kinds of varied-duration or imbalance medical signal processing problems such as Electroencephalogram (EEG) and Photoplethysmography (PPG).

Implementation of a convolutional neural network using Tensorflow machine learning platform

The generalized algorithm of a typical convolutional neural network realization by means of TensorFlow machine learning library is considered. The peculiarities of the coding implementation of the convolutional neural network in the image recognition problem are analyzed with the example of the MNIST datаset.

Slow-Varying Dynamics-Assisted Temporal Capsule Network for Machinery Remaining Useful Life Estimation.

Capsule network (CapsNet) acts as a promising alternative to the typical convolutional neural network, which is the dominant network to develop the remaining useful life (RUL) estimation models for mechanical equipment. Although CapsNet comes with an impressive ability to represent entities' hierarchical relationships through a high-dimensional vector embedding, it fails to capture the long-term temporal correlation of run-to-failure time series measured from degraded mechanical equipment. On the other hand, the slow-varying dynamics, which reveals the low-frequency information hidden in mechanical dynamical behavior, is overlooked in the existing RUL estimation models (including CapsNet), limiting the utmost ability of advanced networks. To address the aforementioned concerns, we propose a slow-varying dynamics-assisted temporal CapsNet (SD-TemCapsNet) to simultaneously learn the slow-varying dynamics and temporal dynamics from measurements for accurate RUL estimation. First, in light of the sensitivity of fault evolution, slow-varying features are decomposed from normal raw data to convey the low-frequency components corresponding to the system dynamics. Next, the long short-term memory (LSTM) mechanism is introduced into CapsNet to capture the temporal correlation of time series. To this end, experiments conducted on an aircraft engine and a milling machine verify that the proposed SD-TemCapsNet outperforms the mainstream methods. In comparison with CapsNet, the estimation accuracy of the aircraft engine with four different scenarios has been improved by 10.17%, 24.97%, 3.25%, and 13.03% about the index root mean squared error, respectively. Similarly, the estimation accuracy of the milling machine has been improved by 23.57% compared to LSTM and 19.54% compared to CapsNet.