Sporadic E(Es)layers in the ionosphere are characterized by intense plasma irregularities in the E region at altitudes of 90-130 km.Because they can significantly influence radio communications and navigation systems,...Sporadic E(Es)layers in the ionosphere are characterized by intense plasma irregularities in the E region at altitudes of 90-130 km.Because they can significantly influence radio communications and navigation systems,accurate forecasting of Es layers is crucial for ensuring the precision and dependability of navigation satellite systems.In this study,we present Es predictions made by an empirical model and by a deep learning model,and analyze their differences comprehensively by comparing the model predictions to satellite RO measurements and ground-based ionosonde observations.The deep learning model exhibited significantly better performance,as indicated by its high coefficient of correlation(r=0.87)with RO observations and predictions,than did the empirical model(r=0.53).This study highlights the importance of integrating artificial intelligence technology into ionosphere modelling generally,and into predicting Es layer occurrences and characteristics,in particular.展开更多
It is fundamental and useful to investigate how deep learning forecasting models(DLMs)perform compared to operational oceanography forecast systems(OFSs).However,few studies have intercompared their performances using...It is fundamental and useful to investigate how deep learning forecasting models(DLMs)perform compared to operational oceanography forecast systems(OFSs).However,few studies have intercompared their performances using an identical reference.In this study,three physically reasonable DLMs are implemented for the forecasting of the sea surface temperature(SST),sea level anomaly(SLA),and sea surface velocity in the South China Sea.The DLMs are validated against both the testing dataset and the“OceanPredict”Class 4 dataset.Results show that the DLMs'RMSEs against the latter increase by 44%,245%,302%,and 109%for SST,SLA,current speed,and direction,respectively,compared to those against the former.Therefore,different references have significant influences on the validation,and it is necessary to use an identical and independent reference to intercompare the DLMs and OFSs.Against the Class 4 dataset,the DLMs present significantly better performance for SLA than the OFSs,and slightly better performances for other variables.The error patterns of the DLMs and OFSs show a high degree of similarity,which is reasonable from the viewpoint of predictability,facilitating further applications of the DLMs.For extreme events,the DLMs and OFSs both present large but similar forecast errors for SLA and current speed,while the DLMs are likely to give larger errors for SST and current direction.This study provides an evaluation of the forecast skills of commonly used DLMs and provides an example to objectively intercompare different DLMs.展开更多
This paper aims to conduct a systematic literature review(SLR)using an artificial intelligence(AI)approach to predict and diagnose diabetes mellitus.After reviewing the literature published from 2015–2025,the paper a...This paper aims to conduct a systematic literature review(SLR)using an artificial intelligence(AI)approach to predict and diagnose diabetes mellitus.After reviewing the literature published from 2015–2025,the paper aims to identify the most effective AI techniques,the most used datasets,the most widely used data preprocessing techniques,and the most common issues.After analyzing the literature,it has been found that convolutional neural networks(CNNs)and long short-term memory(LSTM)networks are deep learning models that have shown high accuracy in diabetes prediction.Recursive feature elimination(RFE)and SMOTE are feature selection techniques that have significantly improved model accuracy,training time,and interpretability.Amidst this technological advancement,some existing issues persist:data imbalance,the inapplicability of techniques,computational limitations,and a lack of real-time application in a healthcare environment.The literature review has also identified the need for robust,interpretable,and scalable AI systems capable of handling large volumes of data,including real-world data,in the healthcare industry.Furthermore,it has been identified that the benefits should be integrated with wearable health monitoring systems and the development of privacy-preserving models to ensure continuous,secure,and proactive diabetes management.展开更多
Accurate forecasting of tropical cyclone(TC)tracks and intensities is essential.Although the TianXing large weather model,a six-hourly forecasting model surpassing operational forecasts,exhibits superior performance,i...Accurate forecasting of tropical cyclone(TC)tracks and intensities is essential.Although the TianXing large weather model,a six-hourly forecasting model surpassing operational forecasts,exhibits superior performance,its TC forecasts still require enhancement.Prediction errors persist due to biases in the training data and smoothing effects in data-driven methods.To address this,we introduce CycloneBCNet,a deep-learning model designed to correct TianXing’s TC forecast biases by leveraging spatial and temporal data.CycloneBCNet utilizes the SimVP(simpler yet better video prediction)framework with spatial attention to highlight cyclone core regions in forecast fields.It also incorporates TC trend information(center position,maximum wind speed,and minimum sea level pressure)via an LSTM(long short-term memory)module.These TC vectors are derived from post-processed TianXing forecasts.By fusing features from forecast fields and TC vectors,CycloneBCNet corrects biases across multiple lead times.At a 96-h lead time,the track error reduces from 162.4 to 86.4 km,the wind speed error from 17.2 to 6.69 m s^(-1),and the pressure error from 22.2 to 9.36 hPa.Interpretability analysis shows that CycloneBCNet adjusts its attention across forecast lead times.Intensity corrections prioritize inner-core dynamics,particularly the eye and eyewall,while track corrections shift from lower-level variables and the cyclone’s core to broader environmental factors and mid-to upper-level features as the forecast duration increases.These findings demonstrate that CycloneBCNet effectively captures key TC dynamics consistent with meteorological principles,including the dominance of near-surface conditions for intensity and the increasing influence of steering currents on track prediction.展开更多
This reviewpresents a comprehensive technical analysis of deep learning(DL)methodologies in biomedical signal processing,focusing on architectural innovations,experimental validation,and evaluation frameworks.We syste...This reviewpresents a comprehensive technical analysis of deep learning(DL)methodologies in biomedical signal processing,focusing on architectural innovations,experimental validation,and evaluation frameworks.We systematically evaluate key deep learning architectures including convolutional neural networks(CNNs),recurrent neural networks(RNNs),transformer-based models,and hybrid systems across critical tasks such as arrhythmia classification,seizure detection,and anomaly segmentation.The study dissects preprocessing techniques(e.g.,wavelet denoising,spectral normalization)and feature extraction strategies(time-frequency analysis,attention mechanisms),demonstrating their impact on model accuracy,noise robustness,and computational efficiency.Experimental results underscore the superiority of deep learning over traditional methods,particularly in automated feature extraction,real-time processing,cross-modal generalization,and achieving up to a 15%increase in classification accuracy and enhanced noise resilience across electrocardiogram(ECG),electroencephalogram(EEG),and electromyogram(EMG)signals.Performance is rigorously benchmarked using precision,recall,F1-scores,area under the receiver operating characteristic curve(AUC-ROC),and computational complexitymetrics,providing a unified framework for comparing model efficacy.Thesurvey addresses persistent challenges:synthetic data generationmitigates limited training samples,interpretability tools(e.g.,Gradient-weighted Class Activation Mapping(Grad-CAM),Shapley values)resolve model opacity,and federated learning ensures privacy-compliant deployments.Distinguished from prior reviews,this work offers a structured taxonomy of deep learning architectures,integrates emerging paradigms like transformers and domain-specific attention mechanisms,and evaluates preprocessing pipelines for spectral-temporal trade-offs.It advances the field by bridging technical advancements with clinical needs,such as scalability in real-world settings(e.g.,wearable devices)and regulatory alignment with theHealth Insurance Portability and Accountability Act(HIPAA)and General Data Protection Regulation(GDPR).By synthesizing technical rigor,ethical considerations,and actionable guidelines for model selection,this survey establishes a holistic reference for developing robust,interpretable biomedical artificial intelligence(AI)systems,accelerating their translation into personalized and equitable healthcare solutions.展开更多
Leveraging the power of artificial intelligence to facilitate an automatic analysis and monitoring of heart sounds has increasingly attracted tremendous efforts in the past decade.Nevertheless,lacking on standard open...Leveraging the power of artificial intelligence to facilitate an automatic analysis and monitoring of heart sounds has increasingly attracted tremendous efforts in the past decade.Nevertheless,lacking on standard open-access database made it difficult to maintain a sustainable and comparable research before the first release of the PhysioNet CinC Challenge Dataset.However,inconsistent standards on data collection,annotation,and partition are still restraining a fair and efficient comparison between different works.To this line,we introduced and benchmarked a first version of the Heart Sounds Shenzhen(HSS)corpus.Motivated and inspired by the previous works based on HSS,we redefined the tasks and make a comprehensive investigation on shallow and deep models in this study.First,we segmented the heart sound recording into shorter recordings(10 s),which makes it more similar to the human auscultation case.Second,we redefined the classification tasks.Besides using the 3 class categories(normal,moderate,and mild/severe)adopted in HSS,we added a binary classification task in this study,i.e.,normal and abnormal.In this work,we provided detailed benchmarks based on both the classic machine learning and the state-of-the-art deep learning technologies,which are reproducible by using open-source toolkits.Last but not least,we analyzed the feature contributions of best performance achieved by the benchmark to make the results more convincing and interpretable.展开更多
This study employs nine distinct deep learning models to categorize 12,444 blood cell images and automatically extract from them relevant information with an accuracy that is beyond that achievable with traditional te...This study employs nine distinct deep learning models to categorize 12,444 blood cell images and automatically extract from them relevant information with an accuracy that is beyond that achievable with traditional techniques.The work is intended to improve current methods for the assessment of human health through measurement of the distribution of four types of blood cells,namely,eosinophils,neutrophils,monocytes,and lymphocytes,known for their relationship with human body damage,inflammatory regions,and organ illnesses,in particular,and with the health of the immune system and other hazards,such as cardiovascular disease or infections,more in general.The results of the experiments show that the deep learning models can automatically extract features from the blood cell images and properly classify them with an accuracy of 98%,97%,and 89%,respectively,with regard to the training,verification,and testing of the corresponding datasets.展开更多
Action recognition is an important topic in computer vision. Recently, deep learning technologies have been successfully used in lots of applications including video data for sloving recognition problems. However, mos...Action recognition is an important topic in computer vision. Recently, deep learning technologies have been successfully used in lots of applications including video data for sloving recognition problems. However, most existing deep learning based recognition frameworks are not optimized for action in the surveillance videos. In this paper, we propose a novel method to deal with the recognition of different types of actions in outdoor surveillance videos. The proposed method first introduces motion compensation to improve the detection of human target. Then, it uses three different types of deep models with single and sequenced images as inputs for the recognition of different types of actions. Finally, predictions from different models are fused with a linear model. Experimental results show that the proposed method works well on the real surveillance videos.展开更多
Background Deep 3D morphable models(deep 3DMMs)play an essential role in computer vision.They are used in facial synthesis,compression,reconstruction and animation,avatar creation,virtual try-on,facial recognition sys...Background Deep 3D morphable models(deep 3DMMs)play an essential role in computer vision.They are used in facial synthesis,compression,reconstruction and animation,avatar creation,virtual try-on,facial recognition systems and medical imaging.These applications require high spatial and perceptual quality of synthesised meshes.Despite their significance,these models have not been compared with different mesh representations and evaluated jointly with point-wise distance and perceptual metrics.Methods We compare the influence of different mesh representation features to various deep 3DMMs on spatial and perceptual fidelity of the reconstructed meshes.This paper proves the hypothesis that building deep 3DMMs from meshes represented with global representations leads to lower spatial reconstruction error measured with L_(1) and L_(2) norm metrics and underperforms on perceptual metrics.In contrast,using differential mesh representations which describe differential surface properties yields lower perceptual FMPD and DAME and higher spatial fidelity error.The influence of mesh feature normalisation and standardisation is also compared and analysed from perceptual and spatial fidelity perspectives.Results The results presented in this paper provide guidance in selecting mesh representations to build deep 3DMMs accordingly to spatial and perceptual quality objectives and propose combinations of mesh representations and deep 3DMMs which improve either perceptual or spatial fidelity of existing methods.展开更多
Photovoltaic(PV)power forecasting is essential for balancing energy supply and demand in renewable energy systems.However,the performance of PV panels varies across different technologies due to differences in efficie...Photovoltaic(PV)power forecasting is essential for balancing energy supply and demand in renewable energy systems.However,the performance of PV panels varies across different technologies due to differences in efficiency and how they process solar radiation.This study evaluates the effectiveness of deep learning models in predicting PV power generation for three panel technologies:Hybrid-Si,Mono-Si,and Poly-Si,across three forecasting horizons:1-step,12-step,and 24-step.Among the tested models,the Convolutional Neural Network—Long Short-Term Memory(CNN-LSTM)architecture exhibited superior performance,particularly for the 24-step horizon,achieving R^(2)=0.9793 and MAE 0.0162 for the Poly-Si array,followed by Mono-Si(R^(2)=0.9768)and Hybrid-Si arrays(R^(2)=0.9769).These findings demonstrate that the CNN-LSTM model can provide accurate and reliable PV power predictions for all studied technologies.By identifying the most suitable predictive model for each panel technology,this study contributes to optimizing PV power forecasting and improving energy management strategies.展开更多
Recent studies have demonstrated a growing global interest in utilising agricultural waste to remediate wastewater.This stems from growing apprehensions about high levels of heavy metals,especially Pb^(2+)ions,in wast...Recent studies have demonstrated a growing global interest in utilising agricultural waste to remediate wastewater.This stems from growing apprehensions about high levels of heavy metals,especially Pb^(2+)ions,in wastewater produced by industrial processes such as mining,paint production,oil refining,smelting,and electroplating.This study examined apple pomace’s Pb^(2+)ions adsorption from wastewater.Response Surface Methodology(RSM)was employed,utilising the central composite face-centred design(CCFD)with three variables:initial concentration(1-50 mg/L),adsorbent dosage(0.1-1 g),and particle size(75-425μm)to formulate a mathematical model for the biosorption of Pb^(2+)ions on apple pomace.An artificial neural network(ANN)was developed using data generated from the RSM design.The CCFD and ANN models showed considerable efficacy in the adsorption process,exhibiting correlation coefficient values of 0.9921 and 0.9999,respectively.The isotherm and kinetic studies were performed,and the Freundlich Isotherm model best fitted the equilibrium data,with a correlation coefficient of 0.972 and a qe of 5.145 mg/g.Additionally,the pseudo-second-order model proved to be the most appropriate for the kinetic data,with an R^(2)of 0.9996.These results confirm that apple pomace functions as an effective,low-cost,and environmentally and sustainably biosorbent for the removal of Pb^(2+)ions from wastewater.Both RSM and ANN models exhibited high predictive capability for the biosorption process.While ANN provides more flexibility in modelling complex non-linear relationships,it is prone to overfitting,particularly with limited datasets,and this was addressed through a 5-fold cross-validation technique.展开更多
Accurate Global Horizontal Irradiance(GHI)forecasting has become vital for successfully integrating solar energy into the electrical grid because of the expanding demand for green power and the worldwide shift favouri...Accurate Global Horizontal Irradiance(GHI)forecasting has become vital for successfully integrating solar energy into the electrical grid because of the expanding demand for green power and the worldwide shift favouring green energy resources.Particularly considering the implications of the aggressive GHG emission targets,accurate GHI forecasting has become vital for developing,designing,and operational managing solar energy systems.This research presented the core concepts of modelling and performance analysis of the application of various forecasting models such as ARIMA(Autoregressive Integrated Moving Average),Elaman NN(Elman Neural Network),RBFN(Radial Basis Function Neural Network),SVM(Support Vector Machine),LSTM(Long Short-Term Memory),Persistent,BPN(Back Propagation Neural Network),MLP(Multilayer Perceptron Neural Network),RF(Random Forest),and XGBoost(eXtreme Gradient Boosting)for assessing multi-seasonal forecasting of GHI.Used the India region data to evaluate the models’performance and forecasting ability.Research using forecasting models for seasonal Global Horizontal Irradiance(GHI)forecasting in winter,spring,summer,monsoon,and autumn.Substantiated performance effectiveness through evaluation metrics,such as Mean Absolute Error(MAE),Root Mean Squared Error(RMSE),and R-squared(R^(2)),coded using Python programming.The performance experimentation analysis inferred that the most accurate forecasts in all the seasons compared to the other forecasting models the Random Forest and eXtreme Gradient Boosting,are the superior and competing models that yield Winter season-based forecasting XGBoost is the best forecasting model with MAE:1.6325,RMSE:4.8338,and R^(2):0.9998.Spring season-based forecasting XGBoost is the best forecasting model with MAE:2.599599,RMSE:5.58539,and R^(2):0.999784.Summer season-based forecasting RF is the best forecasting model with MAE:1.03843,RMSE:2.116325,and R^(2):0.999967.Monsoon season-based forecasting RF is the best forecasting model with MAE:0.892385,RMSE:2.417587,and R^(2):0.999942.Autumn season-based forecasting RF is the best forecasting model with MAE:0.810462,RMSE:1.928215,and R^(2):0.999958.Based on seasonal variations and computing constraints,the findings enable energy system operators to make helpful recommendations for choosing the most effective forecasting models.展开更多
Deep learning(DL),derived from the domain of Artificial Neural Networks(ANN),forms one of the most essential components of modern deep learning algorithms.DL segmentation models rely on layer-by-layer convolution-base...Deep learning(DL),derived from the domain of Artificial Neural Networks(ANN),forms one of the most essential components of modern deep learning algorithms.DL segmentation models rely on layer-by-layer convolution-based feature representation,guided by forward and backward propagation.Acritical aspect of this process is the selection of an appropriate activation function(AF)to ensure robustmodel learning.However,existing activation functions often fail to effectively address the vanishing gradient problem or are complicated by the need for manual parameter tuning.Most current research on activation function design focuses on classification tasks using natural image datasets such asMNIST,CIFAR-10,and CIFAR-100.To address this gap,this study proposesMed-ReLU,a novel activation function specifically designed for medical image segmentation.Med-ReLU prevents deep learning models fromsuffering dead neurons or vanishing gradient issues.It is a hybrid activation function that combines the properties of ReLU and Softsign.For positive inputs,Med-ReLU adopts the linear behavior of ReLU to avoid vanishing gradients,while for negative inputs,it exhibits the Softsign’s polynomial convergence,ensuring robust training and avoiding inactive neurons across the training set.The training performance and segmentation accuracy ofMed-ReLU have been thoroughly evaluated,demonstrating stable learning behavior and resistance to overfitting.It consistently outperforms state-of-the-art activation functions inmedical image segmentation tasks.Designed as a parameter-free function,Med-ReLU is simple to implement in complex deep learning architectures,and its effectiveness spans various neural network models and anomaly detection scenarios.展开更多
The rapid advancement of deep learning and the emergence of largescale neural models,such as bidirectional encoder representations from transformers(BERT),generative pre-trained transformer(GPT),and large language mod...The rapid advancement of deep learning and the emergence of largescale neural models,such as bidirectional encoder representations from transformers(BERT),generative pre-trained transformer(GPT),and large language model Meta AI(LLaMa),have brought significant computational and energy challenges.Neuromorphic computing presents a biologically inspired approach to addressing these issues,leveraging event-driven processing and in-memory computation for enhanced energy efficiency.This survey explores the intersection of neuromorphic computing and large-scale deep learning models,focusing on neuromorphic models,learning methods,and hardware.We highlight transferable techniques from deep learning to neuromorphic computing and examine the memoryrelated scalability limitations of current neuromorphic systems.Furthermore,we identify potential directions to enable neuromorphic systems to meet the growing demands of modern AI workloads.展开更多
Dear Editor,This letter proposes a deep synchronization control(DSC) method to synchronize grid-forming converters with power grids. The method involves constructing a novel controller for grid-forming converters base...Dear Editor,This letter proposes a deep synchronization control(DSC) method to synchronize grid-forming converters with power grids. The method involves constructing a novel controller for grid-forming converters based on the stable deep dynamics model. To enhance the performance of the controller, the dynamics model is optimized within the deep reinforcement learning(DRL) framework. Simulation results verify that the proposed method can reduce frequency deviation and improve active power responses.展开更多
Scene recognition is a critical component of computer vision,powering applications from autonomous vehicles to surveillance systems.However,its development is often constrained by a heavy reliance on large,expensively...Scene recognition is a critical component of computer vision,powering applications from autonomous vehicles to surveillance systems.However,its development is often constrained by a heavy reliance on large,expensively annotated datasets.This research presents a novel,efficient approach that leveragesmulti-model transfer learning from pre-trained deep neural networks—specifically DenseNet201 and Visual Geometry Group(VGG)—to overcome this limitation.Ourmethod significantly reduces dependency on vast labeled data while achieving high accuracy.Evaluated on the Aerial Image Dataset(AID)dataset,the model attained a validation accuracy of 93.6%with a loss of 0.35,demonstrating robust performance with minimal training data.These results underscore the viability of our approach for real-time,data-efficient scene recognition,offering a practical and cost-effective advancement for the field.展开更多
This study aimed to enhance the performance of semantic segmentation for autonomous driving by improving the 2DPASS model.Two novel improvements were proposed and implemented in this paper:dynamically adjusting the lo...This study aimed to enhance the performance of semantic segmentation for autonomous driving by improving the 2DPASS model.Two novel improvements were proposed and implemented in this paper:dynamically adjusting the loss function ratio and integrating an attention mechanism(CBAM).First,the loss function weights were adjusted dynamically.The grid search method is used for deciding the best ratio of 7:3.It gives greater emphasis to the cross-entropy loss,which resulted in better segmentation performance.Second,CBAM was applied at different layers of the 2Dencoder.Heatmap analysis revealed that introducing it after the second block of 2D image encoding produced the most effective enhancement of important feature representation.The training epoch was chosen for optimizing the best value by experiments,which improved model convergence and overall accuracy.To evaluate the proposed approach,experiments were conducted based on the SemanticKITTI database.The results showed that the improved model achieved higher segmentation accuracy by 64.31%,improved 11.47% in mIoU compared with the conventional 2DPASS model(baseline:52.84%).It was more effective at detecting small and distant objects and clearly identifying boundaries between different classes.Issues such as noise and variations in data distribution affected its accuracy,indicating the need for further refinement.Overall,the proposed improvements to the 2DPASS model demonstrated the potential to advance semantic segmentation technology and contributed to a more reliable perception of complex,dynamic environments in autonomous vehicles.Accurate segmentation enhances the vehicle’s ability to distinguish different objects,and this improvement directly supports safer navigation,robust decision-making,and efficient path planning,making it highly applicable to real-world deployment of autonomous systems in urban and highway settings.展开更多
Dear Editor,The letter proposes a tensor low-rank orthogonal compression(TLOC)model for a convolutional neural network(CNN),which facilitates its efficient and highly-accurate low-rank representation.Model compression...Dear Editor,The letter proposes a tensor low-rank orthogonal compression(TLOC)model for a convolutional neural network(CNN),which facilitates its efficient and highly-accurate low-rank representation.Model compression is crucial for deploying deep neural network(DNN)models on resource-constrained embedded devices.展开更多
The paper proposes a new deep structure model,called Densely Connected Cascade Forest-Weighted K Nearest Neighbors(DCCF-WKNNs),to implement the corrosion data modelling and corrosion knowledgemining.Firstly,we collect...The paper proposes a new deep structure model,called Densely Connected Cascade Forest-Weighted K Nearest Neighbors(DCCF-WKNNs),to implement the corrosion data modelling and corrosion knowledgemining.Firstly,we collect 409 outdoor atmospheric corrosion samples of low-alloy steels as experiment datasets.Then,we give the proposed methods process,including random forests-K nearest neighbors(RF-WKNNs)and DCCF-WKNNs.Finally,we use the collected datasets to verify the performance of the proposed method.The results show that compared with commonly used and advanced machine-learning algorithms such as artificial neural network(ANN),support vector regression(SVR),random forests(RF),and cascade forests(cForest),the proposed method can obtain the best prediction results.In addition,the method can predict the corrosion rates with variations of any one single environmental variable,like pH,temperature,relative humidity,SO2,rainfall or Cl-.By this way,the threshold of each variable,upon which the corrosion rate may have a large change,can be further obtained.展开更多
The prediction of fundus fluorescein angiography(FFA)images from fundus structural images is a cutting-edge research topic in ophthalmological image processing.Prediction comprises estimating FFA from fundus camera im...The prediction of fundus fluorescein angiography(FFA)images from fundus structural images is a cutting-edge research topic in ophthalmological image processing.Prediction comprises estimating FFA from fundus camera imaging,single-phase FFA from scanning laser ophthalmoscopy(SLO),and three-phase FFA also from SLO.Although many deep learning models are available,a single model can only perform one or two of these prediction tasks.To accomplish three prediction tasks using a unified method,we propose a unified deep learning model for predicting FFA images from fundus structure images using a supervised generative adversarial network.The three prediction tasks are processed as follows:data preparation,network training under FFA supervision,and FFA image prediction from fundus structure images on a test set.By comparing the FFA images predicted by our model,pix2pix,and CycleGAN,we demonstrate the remarkable progress achieved by our proposal.The high performance of our model is validated in terms of the peak signal-to-noise ratio,structural similarity index,and mean squared error.展开更多
基金supported by the Project of Stable Support for Youth Team in Basic Research Field,CAS(grant No.YSBR-018)the National Natural Science Foundation of China(grant Nos.42188101,42130204)+4 种基金the B-type Strategic Priority Program of CAS(grant no.XDB41000000)the National Natural Science Foundation of China(NSFC)Distinguished Overseas Young Talents Program,Innovation Program for Quantum Science and Technology(2021ZD0300301)the Open Research Project of Large Research Infrastructures of CAS-“Study on the interaction between low/mid-latitude atmosphere and ionosphere based on the Chinese Meridian Project”.The project was supported also by the National Key Laboratory of Deep Space Exploration(Grant No.NKLDSE2023A002)the Open Fund of Anhui Provincial Key Laboratory of Intelligent Underground Detection(Grant No.APKLIUD23KF01)the China National Space Administration(CNSA)pre-research Project on Civil Aerospace Technologies No.D010305,D010301.
文摘Sporadic E(Es)layers in the ionosphere are characterized by intense plasma irregularities in the E region at altitudes of 90-130 km.Because they can significantly influence radio communications and navigation systems,accurate forecasting of Es layers is crucial for ensuring the precision and dependability of navigation satellite systems.In this study,we present Es predictions made by an empirical model and by a deep learning model,and analyze their differences comprehensively by comparing the model predictions to satellite RO measurements and ground-based ionosonde observations.The deep learning model exhibited significantly better performance,as indicated by its high coefficient of correlation(r=0.87)with RO observations and predictions,than did the empirical model(r=0.53).This study highlights the importance of integrating artificial intelligence technology into ionosphere modelling generally,and into predicting Es layer occurrences and characteristics,in particular.
基金supported by the National Natural Science Foundation of China(Grant Nos.42375062 and 42275158)the National Key Scientific and Technological Infrastructure project“Earth System Science Numerical Simulator Facility”(EarthLab)the Natural Science Foundation of Gansu Province(Grant No.22JR5RF1080)。
文摘It is fundamental and useful to investigate how deep learning forecasting models(DLMs)perform compared to operational oceanography forecast systems(OFSs).However,few studies have intercompared their performances using an identical reference.In this study,three physically reasonable DLMs are implemented for the forecasting of the sea surface temperature(SST),sea level anomaly(SLA),and sea surface velocity in the South China Sea.The DLMs are validated against both the testing dataset and the“OceanPredict”Class 4 dataset.Results show that the DLMs'RMSEs against the latter increase by 44%,245%,302%,and 109%for SST,SLA,current speed,and direction,respectively,compared to those against the former.Therefore,different references have significant influences on the validation,and it is necessary to use an identical and independent reference to intercompare the DLMs and OFSs.Against the Class 4 dataset,the DLMs present significantly better performance for SLA than the OFSs,and slightly better performances for other variables.The error patterns of the DLMs and OFSs show a high degree of similarity,which is reasonable from the viewpoint of predictability,facilitating further applications of the DLMs.For extreme events,the DLMs and OFSs both present large but similar forecast errors for SLA and current speed,while the DLMs are likely to give larger errors for SST and current direction.This study provides an evaluation of the forecast skills of commonly used DLMs and provides an example to objectively intercompare different DLMs.
文摘This paper aims to conduct a systematic literature review(SLR)using an artificial intelligence(AI)approach to predict and diagnose diabetes mellitus.After reviewing the literature published from 2015–2025,the paper aims to identify the most effective AI techniques,the most used datasets,the most widely used data preprocessing techniques,and the most common issues.After analyzing the literature,it has been found that convolutional neural networks(CNNs)and long short-term memory(LSTM)networks are deep learning models that have shown high accuracy in diabetes prediction.Recursive feature elimination(RFE)and SMOTE are feature selection techniques that have significantly improved model accuracy,training time,and interpretability.Amidst this technological advancement,some existing issues persist:data imbalance,the inapplicability of techniques,computational limitations,and a lack of real-time application in a healthcare environment.The literature review has also identified the need for robust,interpretable,and scalable AI systems capable of handling large volumes of data,including real-world data,in the healthcare industry.Furthermore,it has been identified that the benefits should be integrated with wearable health monitoring systems and the development of privacy-preserving models to ensure continuous,secure,and proactive diabetes management.
基金supported by the Meteorological Joint Funds of the National Natural Science Foundation of China(Grant No.U2142211)the National Natural Science Foundation of China(Grant Nos.42075141,42341202 and 62088101)+1 种基金the National Key Research and Development Program of China(Grant No.2020YFA0608000)the Shanghai Municipal Science and Technology Major Project(Grant No.2021SHZDZX0100).
文摘Accurate forecasting of tropical cyclone(TC)tracks and intensities is essential.Although the TianXing large weather model,a six-hourly forecasting model surpassing operational forecasts,exhibits superior performance,its TC forecasts still require enhancement.Prediction errors persist due to biases in the training data and smoothing effects in data-driven methods.To address this,we introduce CycloneBCNet,a deep-learning model designed to correct TianXing’s TC forecast biases by leveraging spatial and temporal data.CycloneBCNet utilizes the SimVP(simpler yet better video prediction)framework with spatial attention to highlight cyclone core regions in forecast fields.It also incorporates TC trend information(center position,maximum wind speed,and minimum sea level pressure)via an LSTM(long short-term memory)module.These TC vectors are derived from post-processed TianXing forecasts.By fusing features from forecast fields and TC vectors,CycloneBCNet corrects biases across multiple lead times.At a 96-h lead time,the track error reduces from 162.4 to 86.4 km,the wind speed error from 17.2 to 6.69 m s^(-1),and the pressure error from 22.2 to 9.36 hPa.Interpretability analysis shows that CycloneBCNet adjusts its attention across forecast lead times.Intensity corrections prioritize inner-core dynamics,particularly the eye and eyewall,while track corrections shift from lower-level variables and the cyclone’s core to broader environmental factors and mid-to upper-level features as the forecast duration increases.These findings demonstrate that CycloneBCNet effectively captures key TC dynamics consistent with meteorological principles,including the dominance of near-surface conditions for intensity and the increasing influence of steering currents on track prediction.
基金The Natural Sciences and Engineering Research Council of Canada(NSERC)funded this review study.
文摘This reviewpresents a comprehensive technical analysis of deep learning(DL)methodologies in biomedical signal processing,focusing on architectural innovations,experimental validation,and evaluation frameworks.We systematically evaluate key deep learning architectures including convolutional neural networks(CNNs),recurrent neural networks(RNNs),transformer-based models,and hybrid systems across critical tasks such as arrhythmia classification,seizure detection,and anomaly segmentation.The study dissects preprocessing techniques(e.g.,wavelet denoising,spectral normalization)and feature extraction strategies(time-frequency analysis,attention mechanisms),demonstrating their impact on model accuracy,noise robustness,and computational efficiency.Experimental results underscore the superiority of deep learning over traditional methods,particularly in automated feature extraction,real-time processing,cross-modal generalization,and achieving up to a 15%increase in classification accuracy and enhanced noise resilience across electrocardiogram(ECG),electroencephalogram(EEG),and electromyogram(EMG)signals.Performance is rigorously benchmarked using precision,recall,F1-scores,area under the receiver operating characteristic curve(AUC-ROC),and computational complexitymetrics,providing a unified framework for comparing model efficacy.Thesurvey addresses persistent challenges:synthetic data generationmitigates limited training samples,interpretability tools(e.g.,Gradient-weighted Class Activation Mapping(Grad-CAM),Shapley values)resolve model opacity,and federated learning ensures privacy-compliant deployments.Distinguished from prior reviews,this work offers a structured taxonomy of deep learning architectures,integrates emerging paradigms like transformers and domain-specific attention mechanisms,and evaluates preprocessing pipelines for spectral-temporal trade-offs.It advances the field by bridging technical advancements with clinical needs,such as scalability in real-world settings(e.g.,wearable devices)and regulatory alignment with theHealth Insurance Portability and Accountability Act(HIPAA)and General Data Protection Regulation(GDPR).By synthesizing technical rigor,ethical considerations,and actionable guidelines for model selection,this survey establishes a holistic reference for developing robust,interpretable biomedical artificial intelligence(AI)systems,accelerating their translation into personalized and equitable healthcare solutions.
基金partially supported by the Ministry of Science and Technology of the People's Republic of China with the STI2030-Major Projects(2021ZD0201900)the National Natural Science Foundation of China(No.62227807 and 62272044)+3 种基金the Teli Young Fellow Program from the Beijing Institute of Technology,Chinathe Natural Science Foundation of Shenzhen University General Hospital(No.SUGH2018QD013),Chinathe Shenzhen Science and Technology Innovation Commission Project(No.JCYJ20190808120613189),Chinathe Grants-in-Aid for Scientific Research(No.20H00569)from the Ministry of Education,Culture,Sports,Science and Technology(MEXT),Japan.
文摘Leveraging the power of artificial intelligence to facilitate an automatic analysis and monitoring of heart sounds has increasingly attracted tremendous efforts in the past decade.Nevertheless,lacking on standard open-access database made it difficult to maintain a sustainable and comparable research before the first release of the PhysioNet CinC Challenge Dataset.However,inconsistent standards on data collection,annotation,and partition are still restraining a fair and efficient comparison between different works.To this line,we introduced and benchmarked a first version of the Heart Sounds Shenzhen(HSS)corpus.Motivated and inspired by the previous works based on HSS,we redefined the tasks and make a comprehensive investigation on shallow and deep models in this study.First,we segmented the heart sound recording into shorter recordings(10 s),which makes it more similar to the human auscultation case.Second,we redefined the classification tasks.Besides using the 3 class categories(normal,moderate,and mild/severe)adopted in HSS,we added a binary classification task in this study,i.e.,normal and abnormal.In this work,we provided detailed benchmarks based on both the classic machine learning and the state-of-the-art deep learning technologies,which are reproducible by using open-source toolkits.Last but not least,we analyzed the feature contributions of best performance achieved by the benchmark to make the results more convincing and interpretable.
基金supported by National Natural Science Foundation of China(NSFC)(Nos.61806087,61902158).
文摘This study employs nine distinct deep learning models to categorize 12,444 blood cell images and automatically extract from them relevant information with an accuracy that is beyond that achievable with traditional techniques.The work is intended to improve current methods for the assessment of human health through measurement of the distribution of four types of blood cells,namely,eosinophils,neutrophils,monocytes,and lymphocytes,known for their relationship with human body damage,inflammatory regions,and organ illnesses,in particular,and with the health of the immune system and other hazards,such as cardiovascular disease or infections,more in general.The results of the experiments show that the deep learning models can automatically extract features from the blood cell images and properly classify them with an accuracy of 98%,97%,and 89%,respectively,with regard to the training,verification,and testing of the corresponding datasets.
文摘Action recognition is an important topic in computer vision. Recently, deep learning technologies have been successfully used in lots of applications including video data for sloving recognition problems. However, most existing deep learning based recognition frameworks are not optimized for action in the surveillance videos. In this paper, we propose a novel method to deal with the recognition of different types of actions in outdoor surveillance videos. The proposed method first introduces motion compensation to improve the detection of human target. Then, it uses three different types of deep models with single and sequenced images as inputs for the recognition of different types of actions. Finally, predictions from different models are fused with a linear model. Experimental results show that the proposed method works well on the real surveillance videos.
基金Supported by the Centre for Digital Entertainment at Bournemouth University by the UK Engineering and Physical Sciences Research Council(EPSRC)EP/L016540/1 and Humain Ltd.
文摘Background Deep 3D morphable models(deep 3DMMs)play an essential role in computer vision.They are used in facial synthesis,compression,reconstruction and animation,avatar creation,virtual try-on,facial recognition systems and medical imaging.These applications require high spatial and perceptual quality of synthesised meshes.Despite their significance,these models have not been compared with different mesh representations and evaluated jointly with point-wise distance and perceptual metrics.Methods We compare the influence of different mesh representation features to various deep 3DMMs on spatial and perceptual fidelity of the reconstructed meshes.This paper proves the hypothesis that building deep 3DMMs from meshes represented with global representations leads to lower spatial reconstruction error measured with L_(1) and L_(2) norm metrics and underperforms on perceptual metrics.In contrast,using differential mesh representations which describe differential surface properties yields lower perceptual FMPD and DAME and higher spatial fidelity error.The influence of mesh feature normalisation and standardisation is also compared and analysed from perceptual and spatial fidelity perspectives.Results The results presented in this paper provide guidance in selecting mesh representations to build deep 3DMMs accordingly to spatial and perceptual quality objectives and propose combinations of mesh representations and deep 3DMMs which improve either perceptual or spatial fidelity of existing methods.
文摘Photovoltaic(PV)power forecasting is essential for balancing energy supply and demand in renewable energy systems.However,the performance of PV panels varies across different technologies due to differences in efficiency and how they process solar radiation.This study evaluates the effectiveness of deep learning models in predicting PV power generation for three panel technologies:Hybrid-Si,Mono-Si,and Poly-Si,across three forecasting horizons:1-step,12-step,and 24-step.Among the tested models,the Convolutional Neural Network—Long Short-Term Memory(CNN-LSTM)architecture exhibited superior performance,particularly for the 24-step horizon,achieving R^(2)=0.9793 and MAE 0.0162 for the Poly-Si array,followed by Mono-Si(R^(2)=0.9768)and Hybrid-Si arrays(R^(2)=0.9769).These findings demonstrate that the CNN-LSTM model can provide accurate and reliable PV power predictions for all studied technologies.By identifying the most suitable predictive model for each panel technology,this study contributes to optimizing PV power forecasting and improving energy management strategies.
基金funding from the National Research Foundation of South Africa[Grant No:PMDS240909267358]。
文摘Recent studies have demonstrated a growing global interest in utilising agricultural waste to remediate wastewater.This stems from growing apprehensions about high levels of heavy metals,especially Pb^(2+)ions,in wastewater produced by industrial processes such as mining,paint production,oil refining,smelting,and electroplating.This study examined apple pomace’s Pb^(2+)ions adsorption from wastewater.Response Surface Methodology(RSM)was employed,utilising the central composite face-centred design(CCFD)with three variables:initial concentration(1-50 mg/L),adsorbent dosage(0.1-1 g),and particle size(75-425μm)to formulate a mathematical model for the biosorption of Pb^(2+)ions on apple pomace.An artificial neural network(ANN)was developed using data generated from the RSM design.The CCFD and ANN models showed considerable efficacy in the adsorption process,exhibiting correlation coefficient values of 0.9921 and 0.9999,respectively.The isotherm and kinetic studies were performed,and the Freundlich Isotherm model best fitted the equilibrium data,with a correlation coefficient of 0.972 and a qe of 5.145 mg/g.Additionally,the pseudo-second-order model proved to be the most appropriate for the kinetic data,with an R^(2)of 0.9996.These results confirm that apple pomace functions as an effective,low-cost,and environmentally and sustainably biosorbent for the removal of Pb^(2+)ions from wastewater.Both RSM and ANN models exhibited high predictive capability for the biosorption process.While ANN provides more flexibility in modelling complex non-linear relationships,it is prone to overfitting,particularly with limited datasets,and this was addressed through a 5-fold cross-validation technique.
文摘Accurate Global Horizontal Irradiance(GHI)forecasting has become vital for successfully integrating solar energy into the electrical grid because of the expanding demand for green power and the worldwide shift favouring green energy resources.Particularly considering the implications of the aggressive GHG emission targets,accurate GHI forecasting has become vital for developing,designing,and operational managing solar energy systems.This research presented the core concepts of modelling and performance analysis of the application of various forecasting models such as ARIMA(Autoregressive Integrated Moving Average),Elaman NN(Elman Neural Network),RBFN(Radial Basis Function Neural Network),SVM(Support Vector Machine),LSTM(Long Short-Term Memory),Persistent,BPN(Back Propagation Neural Network),MLP(Multilayer Perceptron Neural Network),RF(Random Forest),and XGBoost(eXtreme Gradient Boosting)for assessing multi-seasonal forecasting of GHI.Used the India region data to evaluate the models’performance and forecasting ability.Research using forecasting models for seasonal Global Horizontal Irradiance(GHI)forecasting in winter,spring,summer,monsoon,and autumn.Substantiated performance effectiveness through evaluation metrics,such as Mean Absolute Error(MAE),Root Mean Squared Error(RMSE),and R-squared(R^(2)),coded using Python programming.The performance experimentation analysis inferred that the most accurate forecasts in all the seasons compared to the other forecasting models the Random Forest and eXtreme Gradient Boosting,are the superior and competing models that yield Winter season-based forecasting XGBoost is the best forecasting model with MAE:1.6325,RMSE:4.8338,and R^(2):0.9998.Spring season-based forecasting XGBoost is the best forecasting model with MAE:2.599599,RMSE:5.58539,and R^(2):0.999784.Summer season-based forecasting RF is the best forecasting model with MAE:1.03843,RMSE:2.116325,and R^(2):0.999967.Monsoon season-based forecasting RF is the best forecasting model with MAE:0.892385,RMSE:2.417587,and R^(2):0.999942.Autumn season-based forecasting RF is the best forecasting model with MAE:0.810462,RMSE:1.928215,and R^(2):0.999958.Based on seasonal variations and computing constraints,the findings enable energy system operators to make helpful recommendations for choosing the most effective forecasting models.
基金The researchers would like to thank the Deanship of Graduate Studies and Scientific Research at Qassim University for financial support(QU-APC-2025).
文摘Deep learning(DL),derived from the domain of Artificial Neural Networks(ANN),forms one of the most essential components of modern deep learning algorithms.DL segmentation models rely on layer-by-layer convolution-based feature representation,guided by forward and backward propagation.Acritical aspect of this process is the selection of an appropriate activation function(AF)to ensure robustmodel learning.However,existing activation functions often fail to effectively address the vanishing gradient problem or are complicated by the need for manual parameter tuning.Most current research on activation function design focuses on classification tasks using natural image datasets such asMNIST,CIFAR-10,and CIFAR-100.To address this gap,this study proposesMed-ReLU,a novel activation function specifically designed for medical image segmentation.Med-ReLU prevents deep learning models fromsuffering dead neurons or vanishing gradient issues.It is a hybrid activation function that combines the properties of ReLU and Softsign.For positive inputs,Med-ReLU adopts the linear behavior of ReLU to avoid vanishing gradients,while for negative inputs,it exhibits the Softsign’s polynomial convergence,ensuring robust training and avoiding inactive neurons across the training set.The training performance and segmentation accuracy ofMed-ReLU have been thoroughly evaluated,demonstrating stable learning behavior and resistance to overfitting.It consistently outperforms state-of-the-art activation functions inmedical image segmentation tasks.Designed as a parameter-free function,Med-ReLU is simple to implement in complex deep learning architectures,and its effectiveness spans various neural network models and anomaly detection scenarios.
文摘The rapid advancement of deep learning and the emergence of largescale neural models,such as bidirectional encoder representations from transformers(BERT),generative pre-trained transformer(GPT),and large language model Meta AI(LLaMa),have brought significant computational and energy challenges.Neuromorphic computing presents a biologically inspired approach to addressing these issues,leveraging event-driven processing and in-memory computation for enhanced energy efficiency.This survey explores the intersection of neuromorphic computing and large-scale deep learning models,focusing on neuromorphic models,learning methods,and hardware.We highlight transferable techniques from deep learning to neuromorphic computing and examine the memoryrelated scalability limitations of current neuromorphic systems.Furthermore,we identify potential directions to enable neuromorphic systems to meet the growing demands of modern AI workloads.
基金supported in part by the National Natural Science Foundation of China(62033005,62273270)the Natural Science Foundation of Shaanxi Province(2023JC-XJ17)
文摘Dear Editor,This letter proposes a deep synchronization control(DSC) method to synchronize grid-forming converters with power grids. The method involves constructing a novel controller for grid-forming converters based on the stable deep dynamics model. To enhance the performance of the controller, the dynamics model is optimized within the deep reinforcement learning(DRL) framework. Simulation results verify that the proposed method can reduce frequency deviation and improve active power responses.
基金funded by the Deanship of Scientific Research and Libraries,Princess Nourah bint Abdulrahman University,through the Program of Research Project Funding After Publication,grant No.(RPFAP-23-1445).
文摘Scene recognition is a critical component of computer vision,powering applications from autonomous vehicles to surveillance systems.However,its development is often constrained by a heavy reliance on large,expensively annotated datasets.This research presents a novel,efficient approach that leveragesmulti-model transfer learning from pre-trained deep neural networks—specifically DenseNet201 and Visual Geometry Group(VGG)—to overcome this limitation.Ourmethod significantly reduces dependency on vast labeled data while achieving high accuracy.Evaluated on the Aerial Image Dataset(AID)dataset,the model attained a validation accuracy of 93.6%with a loss of 0.35,demonstrating robust performance with minimal training data.These results underscore the viability of our approach for real-time,data-efficient scene recognition,offering a practical and cost-effective advancement for the field.
文摘This study aimed to enhance the performance of semantic segmentation for autonomous driving by improving the 2DPASS model.Two novel improvements were proposed and implemented in this paper:dynamically adjusting the loss function ratio and integrating an attention mechanism(CBAM).First,the loss function weights were adjusted dynamically.The grid search method is used for deciding the best ratio of 7:3.It gives greater emphasis to the cross-entropy loss,which resulted in better segmentation performance.Second,CBAM was applied at different layers of the 2Dencoder.Heatmap analysis revealed that introducing it after the second block of 2D image encoding produced the most effective enhancement of important feature representation.The training epoch was chosen for optimizing the best value by experiments,which improved model convergence and overall accuracy.To evaluate the proposed approach,experiments were conducted based on the SemanticKITTI database.The results showed that the improved model achieved higher segmentation accuracy by 64.31%,improved 11.47% in mIoU compared with the conventional 2DPASS model(baseline:52.84%).It was more effective at detecting small and distant objects and clearly identifying boundaries between different classes.Issues such as noise and variations in data distribution affected its accuracy,indicating the need for further refinement.Overall,the proposed improvements to the 2DPASS model demonstrated the potential to advance semantic segmentation technology and contributed to a more reliable perception of complex,dynamic environments in autonomous vehicles.Accurate segmentation enhances the vehicle’s ability to distinguish different objects,and this improvement directly supports safer navigation,robust decision-making,and efficient path planning,making it highly applicable to real-world deployment of autonomous systems in urban and highway settings.
基金supported by the Science and Technology Innovation Key R&D Program of Chongqing(CSTB2025TIAD-STX0032)National Key Research and Development Program of China(2024YFF0908200)+1 种基金the Chongqing Technology Innovation and Application Development Special Key Project(CSTB2024TIAD-KPX0018)the Southwest University Graduate Student Research Innovation(SWUB24051)。
文摘Dear Editor,The letter proposes a tensor low-rank orthogonal compression(TLOC)model for a convolutional neural network(CNN),which facilitates its efficient and highly-accurate low-rank representation.Model compression is crucial for deploying deep neural network(DNN)models on resource-constrained embedded devices.
基金financially supported by the National Key R&D Program of China(No.2017YFB0702100)the National Natural Science Foundation of China(No.51871024)。
文摘The paper proposes a new deep structure model,called Densely Connected Cascade Forest-Weighted K Nearest Neighbors(DCCF-WKNNs),to implement the corrosion data modelling and corrosion knowledgemining.Firstly,we collect 409 outdoor atmospheric corrosion samples of low-alloy steels as experiment datasets.Then,we give the proposed methods process,including random forests-K nearest neighbors(RF-WKNNs)and DCCF-WKNNs.Finally,we use the collected datasets to verify the performance of the proposed method.The results show that compared with commonly used and advanced machine-learning algorithms such as artificial neural network(ANN),support vector regression(SVR),random forests(RF),and cascade forests(cForest),the proposed method can obtain the best prediction results.In addition,the method can predict the corrosion rates with variations of any one single environmental variable,like pH,temperature,relative humidity,SO2,rainfall or Cl-.By this way,the threshold of each variable,upon which the corrosion rate may have a large change,can be further obtained.
基金supported in part by the Gusu Innovation and Entrepreneurship Leading Talents in Suzhou City,grant numbers ZXL2021425 and ZXL2022476Doctor of Innovation and Entrepreneurship Program in Jiangsu Province,grant number JSSCBS20211440+6 种基金Jiangsu Province Key R&D Program,grant number BE2019682Natural Science Foundation of Jiangsu Province,grant number BK20200214National Key R&D Program of China,grant number 2017YFB0403701National Natural Science Foundation of China,grant numbers 61605210,61675226,and 62075235Youth Innovation Promotion Association of Chinese Academy of Sciences,grant number 2019320Frontier Science Research Project of the Chinese Academy of Sciences,grant number QYZDB-SSW-JSC03Strategic Priority Research Program of the Chinese Academy of Sciences,grant number XDB02060000.
文摘The prediction of fundus fluorescein angiography(FFA)images from fundus structural images is a cutting-edge research topic in ophthalmological image processing.Prediction comprises estimating FFA from fundus camera imaging,single-phase FFA from scanning laser ophthalmoscopy(SLO),and three-phase FFA also from SLO.Although many deep learning models are available,a single model can only perform one or two of these prediction tasks.To accomplish three prediction tasks using a unified method,we propose a unified deep learning model for predicting FFA images from fundus structure images using a supervised generative adversarial network.The three prediction tasks are processed as follows:data preparation,network training under FFA supervision,and FFA image prediction from fundus structure images on a test set.By comparing the FFA images predicted by our model,pix2pix,and CycleGAN,we demonstrate the remarkable progress achieved by our proposal.The high performance of our model is validated in terms of the peak signal-to-noise ratio,structural similarity index,and mean squared error.
