Combined Deep Learning Research Articles

Short-term regional wind power forecasting based on spatial–temporal correlation and dynamic clustering model

Short-term regional wind power forecasting (WPF) is essential for enhancing the power grid’s robustness. In the field of regional short-term power forecasting, numerical weather forecasts (NWP) and regional wind-power historical data have been commonly used in power forecasting. In this paper, a regional wind power prediction model based on the spatial and temporal correlation of meteorological resources is proposed to predict the wind power in the next seven days. First, combining variational mode decomposition and Granger causality test (VMD-GCT) to analyze the NWP-related meteorological factors to the wind power components in different bands to obtain the screening of the NWP-related factors. Then, the correlation analysis technique, which uses the physical characteristics of the wind speed time lag in a region, dynamically analyzes the spatial correlation time lag between a wind farm in the region and other wind farms, and uses the DBSCAN (density-based spatial clustering of applications with noise) algorithm to dynamically divide wind power clusters in the wind power forecast cycle, thus laying the foundation for the input data of the prediction model. Finally, the short-term power forecasting in the region is performed by a combined deep learning model. The results show that the proposed method can significantly improve the accuracy and efficiency of the wind power prediction for clusters at different wind speeds. The proposed method is of great significance for the short-term prediction of wind power in a region.

Integrating an attention-based deep learning framework and the SAFY-V model for winter wheat yield estimation using time series SAR and optical data

Information on the spatial distribution of yields can be obtained over a large area by using remote sensing (RS) data. Combining Synthetic Aperture Radar (SAR), being sensitive to above ground biomass and soil moisture in all weather conditions, and optical data can improve the usability of RS data and provide a basis for pixel-based crop yield estimation (YE). In this study, an Upscaled Convolutional Gated Recurrent Unit model incorporated an attention mechanism (UpSc-AConvGRU model) was proposed to improve the estimation accuracy of the winter wheat growth parameter, Leaf Area Index (LAI). Gap filling the time series of optical data was done with backscatter coefficients, local incidence angles and polarimetric decomposition information from Sentinel-1 SAR imagery. The time series LAI estimated by the UpSc-AConvGRU model and Vegetation Temperature Condition Index (VTCI) retrieved from Sentinel-3 optical imagery were then used as state variables of the SAFY-V model to estimate winter wheat yield. The results showed that the proposed UpSc-AConvGRU model incorporated the Convolutional Block Attention Module (CBAM) can effectively improve the accuracy of LAI estimation, with RMSEs ranging from 0.413 to 0.699 m2 m2 for LAI estimated within main growth stages (MGSs) of winter wheat. The correlation between estimated LAI and Sentinel-3 retrieved LAI was generally higher at irrigated farmland compared to rain-fed farmland. The estimated LAI was closest to Sentinel-3 retrieved LAI at the green-up and late heading-filling stage of winter wheat, followed by the jointing and early heading-filling stage, and finally the milk maturity stage. There was good agreement between the SAFY-V model estimated and field measured winter wheat yields (R2 = 0.546, RMSE = 0.757 t ha−1), and the estimated yields at the pixel scale in the Guanzhong Plain, PR China were satisfactory. This study combined deep learning and crop growth modeling, proposed a new pixel scale winter wheat YE method.

A Combined Computer Vision and Deep Learning Approach for Rapid Drone-Based Optical Characterization of Parabolic Troughs

Abstract Optical accuracy is a primary driver of parabolic trough concentrating solar power (CSP) plant performance, but can be damaged by wind loads, gravity, error during installation, and regular plant operation. Collecting and analyzing optical measurements over an entire operating parabolic trough plant is difficult, given the large scale of typical installations. Distant Observer, a software tool developed at the National Renewable Energy Laboratory, uses images of the absorber tube reflected in the collector mirror to measure both surface slope in the parabolic mirror and offset of the absorber tube from the ideal focal point. This technology has been adapted for fast data collection using low-cost commercial drones, but until recently still required substantial human labor to process large amounts of data. A new method leveraging advanced deep learning and computer vision tools can drastically reduce the time required to process images. This new method addresses the primary analysis bottleneck, identifying featureless, reflective mirror corner points to a high degree of accuracy. Recent work has shown promising results using computer vision methods. The combined deep learning and computer vision approach presented here proved highly effective and has the potential to further automate data collection and analysis, making the tool more robust. The method presented in this paper automatically identified 74.3% of mirror corners within 2 pixels of their manually marked counterparts and 91.9% within 3 pixels. This level of accuracy is sufficient for practical Distant Observer analysis within a target uncertainty. A commercial drone collected video of over 100 parabolic trough modules at an operating CSP plant to demonstrate the deep learning and computer vision method’s usefulness in processing large amounts of data. These troughs were successfully analyzed using Distant Observer, paired with the new deep learning and computer vision algorithm, and can provide plant operators and trough designers with valuable insight about plant performance, operating strategies, and plant-wide optical error trends.

Luotuo Mountain Waste Dump Cover Interpretation Combining Deep Learning and VDVI Based on Data from an Unmanned Aerial Vehicle (UAV)

Exposed mine gangue hills are prone to environmental problems such as soil erosion, surface water pollution, and dust. Revegetation of gangue hills can effectively combat the problem. Effective ground cover monitoring means can significantly improve the efficiency of vegetation restoration. We used UAV aerial photography to acquire data and used the Real-SR network to reconstruct the data in super-resolution; the Labv3+ network was used to segment the ground cover into green areas, open spaces, roads, and waters, and VDVI and Otsu were used to extract the vegetation from the green areas. The final ground-cover decomposition accuracy of this method can reach 82%. The application of a super-resolution reconstruction network improves the efficiency of UAV aerial photography; the ground interpretation method of deep learning combined with a vegetation index solves both the problem that vegetation index segmentation cannot cope with the complex ground and the problem of low accuracy due to little data for deep-learning image segmentation.

Automated detection of microfossil fish teeth from slide images using combined deep learning models

Microfossil fish teeth, known as ichthyoliths, provide a key constraint on the depositional age and environment of deep-sea sediments, especially pelagic clays where siliceous and calcareous microfossils are rarely observed. However, traditional methods for the observation of ichthyoliths require considerable time and manual labor, which can hinder their wider application. In this study, we constructed a system to automatically detect ichthyoliths in microscopic images by combining two open source deep learning models. First, the regions for ichthyoliths within the microscopic images are predicted by the instance segmentation model Mask R–CNN. All the detected regions are then re-classified using the image classification model EfficientNet-V2 to determine the classes more accurately. Compared with only using the Mask R–CNN model, the combined system offers significantly higher performance (89.0% precision, 78.6% recall, and an F1 score of 83.5%), demonstrating the utility of the system. Our system can also predict the lengths of the teeth that have been detected, with more than 90% of the predicted lengths being within ±20% of measured length. This system provides a novel, automated, and reliable approach for the detection and length measurement of ichthyoliths from microscope images that can be applied in a range of paleoceanographic and paleoecological contexts.

Risk Measurement of the Financial Credit Industry Driven by Data

The risk measurement of financial credit industry is an important research issue in the field of financial risk assessment. The design of financial credit risk measurement algorithm can help investors avoid greater risks and obtain higher returns, so as to promote the benign development of financial credit industry. Based on the combined deep learning algorithm, this paper studies the risk measurement of financial and credit industry, and proposes a fusion algorithm of deep auto-encoder (DAE) and Long Short-Term Memory (LSTM) network. The algorithm recombines the value of fixed features by using the unsupervised mechanism of DAE, and extracts non fixed features for measurement combined with the memory characteristics of LSTM network. The experimental results show that: compared with single generalized regression neural network and LSTM network, the average accuracy of DAE-LSTM algorithm is improved by about 6.49% and 3.25% respectively, which has a better application effect in credit risk measurement.

AI for predicting chemical-effect associations at the chemical universe level-deepFPlearn.

Many chemicals are present in our environment, and all living species are exposed to them. However, numerous chemicals pose risks, such as developing severe diseases, if they occur at the wrong time in the wrong place. For the majority of the chemicals, these risks are not known. Chemical risk assessment and subsequent regulation of use require efficient and systematic strategies. Lab-based methods—even if high throughput—are too slow to keep up with the pace of chemical innovation. Existing computational approaches are designed for specific chemical classes or sub-problems but not usable on a large scale. Further, the application range of these approaches is limited by the low amount of available labeled training data. We present the ready-to-use and stand-alone program deepFPlearn that predicts the association between chemical structures and effects on the gene/pathway level using a combined deep learning approach. deepFPlearn uses a deep autoencoder for feature reduction before training a deep feed-forward neural network to predict the target association. We received good prediction qualities and showed that our feature compression preserves relevant chemical structural information. Using a vast chemical inventory (unlabeled data) as input for the autoencoder did not reduce our prediction quality but allowed capturing a much more comprehensive range of chemical structures. We predict meaningful—experimentally verified—associations of chemicals and effects on unseen data. deepFPlearn classifies hundreds of thousands of chemicals in seconds. We provide deepFPlearn as an open-source and flexible tool that can be easily retrained and customized to different application settings at https://github.com/yigbt/deepFPlearn.

A Deep Transfer Learning Based Architecture for Brain Tumor Classification Using MR Images

Deep Learning (DL) is becoming more popular in the healthcare sectors due to the exponential growth of data availability and its excellent performance in diagnosing various diseases. This paper has aimed to design the best possible brain tumor diagnostic model to improve accuracy and reliability of radiology. In this paper, an advanced deep learning algorithm is used to detect and classify brain tumors in magnetic resonance (MR) images. Diagnosing brain tumors in radiology is a significant issue, yet it is a difficult and time-consuming procedure that radiologists must pass through. The reliability of their assessment relies completely on their knowledge and personal judgements which are in most cases inaccurate. In response to the growing concern about the inaccuracies in the diagnosis of brain tumors in recent years, this paper combined deep learning and radiometric technologies and perfectly classified brain MR images with high performance accuracy. The research leveraged a transfer learning model known as AlexNet's convolutional neural network (CNN) to perform this operation. Our method helps us to improve robustness, efficiencies and accuracy in the healthcare sector with the ability to automate the entire diagnostic process with the overall accuracy of 99.62%. Additionally, our model has the ability to detect and classify tumors at their different stages and magnitudes.

Event Scene Method of Legal Domain Knowledge Map Based on Neural Network Hybrid Model.

Event extraction technology is one of the important researches in the field of information extraction, which helps people accurately retrieve, find, classify, and summarize effective information from a large amount of information streams. This paper uses the neural network hybrid model to identify the trigger words and event categories of the legal domain knowledge graph events, extracts the events of interest from a large amount of free text, and displays them in a structured format. First, the original text is preprocessed, and then, the distributed semantic word vector is combined with the dependent syntactic structure and location attributes to create a semantic representation in the form of a vector. The combined deep learning model is used to extract activated words, the long-term memory loop neural network uses temporal semantics to extract deep features, and the convergent neural network completes the extraction of activated words and event categories. Finally, the experimental results show that the accuracy of event extraction of the neural network hybrid model designed in this paper has reached 77.1%, and the recall rate has reached 76.8%, which is greatly improved compared with the traditional model.

A comprehensible machine learning tool to differentially diagnose idiopathic pulmonary fibrosis from other chronic interstitial lung diseases.

Idiopathic pulmonary fibrosis (IPF) has poor prognosis, and the multidisciplinary diagnostic agreement is low. Moreover, surgical lung biopsies pose comorbidity risks. Therefore, using data from non-invasive tests usually employed to assess interstitial lung diseases (ILDs), we aimed to develop an automated algorithm combining deep learning and machine learning that would be capable of detecting and differentiating IPF from other ILDs. We retrospectively analysed consecutive patients presenting with ILD between April 2007 and July 2017. Deep learning was used for semantic image segmentation of HRCT based on the corresponding labelled images. A diagnostic algorithm was then trained using the semantic results and non-invasive findings. Diagnostic accuracy was assessed using five-fold cross-validation. In total, 646,800 HRCT images and the corresponding labelled images were acquired from 1068 patients with ILD, of whom 42.7% had IPF. The average segmentation accuracy was 96.1%. The machine learning algorithm had an average diagnostic accuracy of 83.6%, with high sensitivity, specificity and kappa coefficient values (80.7%, 85.8% and 0.665, respectively). Using Cox hazard analysis, IPF diagnosed using this algorithm was a significant prognostic factor (hazard ratio, 2.593; 95% CI, 2.069-3.250; p < 0.001). Diagnostic accuracy was good even in patients with usual interstitial pneumonia patterns on HRCT and those with surgical lung biopsies. Using data from non-invasive examinations, the combined deep learning and machine learning algorithm accurately, easily and quickly diagnosed IPF in a population with various ILDs.

Combining Numerical Simulation and Deep Learning for Landslide Displacement Prediction: An Attempt to Expand the Deep Learning Dataset

Effective landslide hazard prevention requires accurate landslide prediction models, and the data-driven approaches based on deep learning models are gradually becoming a hot research topic. When training deep learning models, it is always preferable to have a large dataset, while most available landslide monitoring data are limited. For data missing or data sparseness problems, conventional interpolation methods based on mathematical knowledge lack mechanism interpretability. This paper proposes that numerical simulations can be used to expand the deep learning dataset we need. Taking the Jiuxianping landslide in the Three Gorges Reservoir Area (TGRA) as the geological background, a finite element numerical model was established, and the landslide displacement time series data were solved considering the boundary conditions of reservoir water level change and precipitation. Next, based on three metrics: Euclidean distance, cosine similarity, and dynamic time warping (DTW) distance, the time series similarity between the displacement data obtained from simulation and data obtained from actual monitoring were verified. Finally, the combined deep learning model was built to predict the displacement of the Jiuxianping landslide. The model was trained on both the simulated and monitored datasets and tested by the last 12 monitored data points. Prediction results with the testing set showed that the models trained using the expanded training set from numerical simulations exhibited lower prediction errors, and the errors had a more concentrated distribution. The results suggest that this landslide displacement prediction method combining numerical simulation and deep learning can solve the problem of inadequate datasets due to low monitoring frequency, as well as provide an interpretation of the physical mechanism for data vacancy filling.

A combined deep learning and physical modelling method for estimating air pollutants’ source location and emission profile in street canyons

Roadside air pollution monitoring stations have become frequently available for street canyons. To efficiently estimate source location and emission profile in street canyons, this study developed a combined deep learning and physical modelling method using the monitoring data as inputs. First, a deep neural network (DNN) was constructed for locating the source. The training datasets were generated from numerical simulations by the computational fluid dynamics (CFD)-Markov chain model. An inverse method based on Tikhonov regularization was then used to estimate the emission profile. Finally, the Markov chain model was used to calculate the air pollutant distribution in the whole street canyon. Case studies were conducted to demonstrate the performance of the proposed method. For the unit impulse source in the 2-D ventilated chamber of 27 m2, the source in 83% of the cases were accurately identified, and in another 13% of the cases, the identified source was within 0.4 m to the true location. For the continuous pollutant source with varying emission profile in the 3-D street canyon with an area of 25,600 m2, the source in 36% of the cases were accurately located, and in another 52% of the cases, it was within 10 m from the true location.

A combined deep learning model for damage size estimation of rolling bearing

The identification of damage sizes based on vibration monitoring data is significant in the study of fault evolution, prediction and diagnosis of rolling bearing. There are inherent restrictions in traditional identification models such as high dependence on prior knowledge and insufficient feature extraction, aiming at the limitation, a prediction method of rolling bearing damage size based on deep learning is proposed. In this paper, a combined model of deep convolutional long-short-term memory network is developed, which can sufficiently extract the multi-dimensional and time-series characteristics of bearing vibration signal, and realize the intelligent and efficient diagnosis of bearing fault. On the basis of theoretical analysis, the rolling bearing fault tests under various damage sizes and rotational velocities are carried out by using the accelerated fatigue testing machine for rolling bearings, and the traditional and novel methods are compared based on the test data. The results show that the prediction accuracy of the combined network can reach 99.94% and 98.67%, respectively under normal and noisy conditions, which is higher than the single deep convolution network, long-short-term memory network and other models. The comparison results amply demonstrate the superiority of the proposed method.

Distributed Deep CNN-LSTM Model for Intrusion Detection Method in IoT-Based Vehicles

As 5G and other technologies are widely used in the Internet of Vehicles, intrusion detection plays an increasingly important role as a vital detection tool for information security. However, due to the rapid changes in the structure of the Internet of Vehicles, the large data flow, and the complex and diverse forms of intrusion, traditional detection methods cannot ensure their accuracy and real-time requirements and cannot be directly applied to the Internet of Vehicles. A new AA distributed combined deep learning intrusion detection method for the Internet of Vehicles based on the Apache Spark framework is proposed in response to these problems. The cluster combines deep-learning convolutional neural network (CNN) and extended short-term memory (LSTM) network to extract features and data for detection of car network intrusion from large-scale car network data traffic and discovery of abnormal behavior. The experimental results show that compared with other existing models, the algorithm of this model can reach 20 in the fastest time, and the accuracy rate is up to 99.7%, with a good detection effect.

A contrastive learning approach for\xa0ICU false arrhythmia\xa0alarm reduction

The high rate of false arrhythmia alarms in Intensive Care Units (ICUs) can lead to disruption of care, negatively impacting patients’ health through noise disturbances, and slow staff response time due to alarm fatigue. Prior false-alarm reduction approaches are often rule-based and require hand-crafted features from physiological waveforms as inputs to machine learning classifiers. Despite considerable prior efforts to address the problem, false alarms are a continuing problem in the ICUs. In this work, we present a deep learning framework to automatically learn feature representations of physiological waveforms using convolutional neural networks (CNNs) to discriminate between true vs. false arrhythmia alarms. We use Contrastive Learning to simultaneously minimize a binary cross entropy classification loss and a proposed similarity loss from pair-wise comparisons of waveform segments over time as a discriminative constraint. Furthermore, we augment our deep models with learned embeddings from a rule-based method to leverage prior domain knowledge for each alarm type. We evaluate our method using the dataset from the 2015 PhysioNet Computing in Cardiology Challenge. Ablation analysis demonstrates that Contrastive Learning significantly improves the performance of a combined deep learning and rule-based-embedding approach. Our results indicate that the final proposed deep learning framework achieves superior performance in comparison to the winning entries of the Challenge.

Классификация гиперспектральных данных дистанционного зондирования Земли с использованием комбинированных 3D--2D сверточных нейронных сетей

Hyperspectral image classification is used for analyzing remote Earth sensing data. Convolutional neural network is one of the most commonly used methods for processing visual data based on deep learning. The article considers the proposed hybrid 3D--2D spectral convolutional neural network for hyperspectral image classification. At the initial stage, a simple combined trained deep learning model was proposed, which was constructed by combining 2D and 3D convolutional neural networks to extract deeper spatial-spectral features with fewer 3D--2D convolutions. The 3D network facilitates the joint spatial-spectral representation of objects from a stack of spectral bands. Functions of 3D--2D convolutional neural networks were used for classifying hyperspectral images. The algorithm of the method of principal components is applied to reduce the dimension. Hyperspectral image classification experiments were performed on Indian Pines, University of Pavia and Salinas Scene remote sensing datasets. The first layer of the feature map is used as input for subsequent layers in predicting final labels for each hyperspectral pixel. The proposed method not only includes the benefits of advanced feature extraction from convolutional neural networks, but also makes full use of spectral and spatial information. The effectiveness of the proposed method was tested on three reference data sets. The results show that a multifunctional learning system based on such networks significantly improves classification accuracy (more than 99 %)

A Combined Deep Learning System for Automatic Detection of “Bovine” Aortic Arch on Computed Tomography Scans

The “bovine” aortic arch is an anatomic variant consisting in a common origin of the innominate and left carotid artery (CILCA), associated with a greater risk of thoracic aortic diseases (aneurysms and dissections), stroke, and complications after endovascular procedures. CILCA can be detected by visual assessment of computed tomography (CT) chest scans, but it is rarely reported. We developed a deep learning (DL) segmentation-plus-classification system to automatically detect CILCA based on 302 CT studies acquired at 2 centers. One model (3D U-Net) was trained from scratch (supervised by manual segmentation), validated, and tested for the automatic segmentation of the aortic arch and supra-aortic vessels. Three DL architectures (ResNet50, DenseNet-201, and SqueezeNet), pre-trained over millions of common images, were trained, validated, and tested for the automatic classification of CILCA versus non-CILCA, supervised by radiologist’s classification. The 3D U-Net-plus-DenseNet-201 was found to be the best system (Dice index 0.912); its classification performance obtained from internal, independent testing on 126 patients gave a receiver operating characteristic area under the curve of 87.0%, sensitivity 66.7%, specificity 90.5%, positive predictive value 87.5%, negative predictive value 73.1%, positive likelihood ratio 7.0, and negative likelihood ratio 0.4. In conclusion, a combined DL system applied to chest CT scans was developed and proven to be an effective tool to detect individuals with “bovine” aortic arch with a low rate of false-positive findings.

AB-LaBSE: Uyghur Sentiment Analysis via the Pre-Training Model with BiLSTM

In recent years, more and more attention has been paid to text sentiment analysis, which has gradually become a research hotspot in information extraction, data mining, Natural Language Processing (NLP), and other fields. With the gradual popularization of the Internet, sentiment analysis of Uyghur texts has great research and application value in online public opinion. For low-resource languages, most state-of-the-art systems require tens of thousands of annotated sentences to get high performance. However, there is minimal annotated data available about Uyghur sentiment analysis tasks. There are also specificities in each task—differences in words and word order across languages make it a challenging problem. In this paper, we present an effective solution to providing a meaningful and easy-to-use feature extractor for sentiment analysis tasks: using the pre-trained language model with BiLSTM layer. Firstly, data augmentation is carried out by AEDA (An Easier Data Augmentation), and the augmented dataset is constructed to improve the performance of text classification tasks. Then, a pretraining model LaBSE is used to encode the input data. Then, BiLSTM is used to learn more context information. Finally, the validity of the model is verified via two categories datasets for sentiment analysis and five categories datasets for emotion analysis. We evaluated our approach on two datasets, which showed wonderful performance compared to some strong baselines. We close with an overview of the resources for sentiment analysis tasks and some of the open research questions. Therefore, we propose a combined deep learning and cross-language pretraining model for two low resource expectations.

A combined deep learning model based on the ideal distance weighting method for fake news detection

Fake news has become a major problem affecting people, society, the economy and national security. This work proposes a combined deep learning model based on the ideal distance weighting method for fake news detection. The proposed model was validated on the ISOT and COVID-19 fake news datasets. Firstly, the ISOT and COVID-19 fake news datasets were collected. Secondly, the training-based models were used to provide accuracy values. After that, these values were transformed into criteria weights using the new ideal distance weighting method. Finally, the prediction value of the proposed model is calculated by the criteria weights. The results show that the proposed method is effective to distinguish the fake news datasets.

A combined deep learning load forecasting model of single household resident user considering multi-time scale electricity consumption behavior

With the growth of residential load and the popularity of intelligent devices, resident users have become important target customers for demand response (DR). However, due to the strong volatility of individual household load and the large difference in user’s behavior, the accuracy of residential load forecasting is generally low and the forecasting effect is unstable, which is not conductive to the implementation of DR. To improve the accuracy of residential load forecasting, this paper proposes a combined deep learning load forecasting model considering multi-time scale electricity consumption behavior of single household resident user to achieve high-accuracy and stable load forecasting. Aiming at the electricity consumption behavior, the multi-time scale similarity analysis is carried out. For the time scale of one year, Normalized Dynamic Time Warping (N-DTW) and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) are used to analyze the significance of single user's long-term electricity consumption behavior. For the time scale of 7 days, behavior similarity is used to analyze the consistency of single user's short-term electricity consumption behavior. Then, Mutual Information (MI) and Principal Component Analysis (PCA) are used to select features and reduce dimensions of multi-dimensional weather influencing factors, so as to avoid the interference of irrelevant factors and improve the calculation speed. On this basis, combined with Back Propagation (BP) neural network, Extreme Gradient Boosting (XGBoost) and Long Short-Term Memory (LSTM) neural network, a combined deep learning network load forecasting model (Co-LSTM) is constructed by using multi-model and multi-variable method to achieve stable and high-accuracy load forecasting. Finally, based on the actual load data from the American Pecan Street Energy Project, the forecasting accuracy of the proposed model of resident user is evaluated. From the performance of load forecasting for 42 target users, the minimum, maximum and average Mean Arctangent Absolute Percentage Error (MAAPE) of Co-LSTM is 18.70%, 45.95% and 31.20% (the average MAAPE is 4.97% less than the traditional LSTM model) respectively.