Error Metrics Research Articles

Category-Aware Saliency Enhance Learning Based on CLIP for Weakly Supervised Salient Object Detection

Weakly supervised salient object detection (SOD) using image-level category labels has been proposed to reduce the annotation cost of pixel-level labels. However, existing methods mostly train a classification network to generate a class activation map, which suffers from coarse localization and difficult pseudo-label updating. To address these issues, we propose a novel Category-aware Saliency Enhance Learning (CSEL) method based on contrastive vision-language pre-training (CLIP), which can perform image-text classification and pseudo-label updating simultaneously. Our proposed method transforms image-text classification into pixel-text matching and generates a category-aware saliency map, which is evaluated by the classification accuracy. Moreover, CSEL assesses the quality of the category-aware saliency map and the pseudo saliency map, and uses the quality confidence scores as weights to update the pseudo labels. The two maps mutually enhance each other to guide the pseudo saliency map in the correct direction. Our SOD network can be trained jointly under the supervision of the updated pseudo saliency maps. We test our model on various well-known RGB-D and RGB SOD datasets. Our model achieves an S-measure of 87.6%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document} on the RGB-D NLPR dataset and 84.3%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document} on the RGB ECSSD dataset. Additionally, we obtain satisfactory performance on the weakly supervised E-measure, F-measure, and mean absolute error metrics for other datasets. These results demonstrate the effectiveness of our model.

Enhancing Additive Restoration of Damaged Polymer Curved Surfaces through Compensated Support Beam Utilization

As additive manufacturing advances, it offers a cost-effective avenue for structurally repairing components. However, a challenge arises in the additive repair of suspended damaged surfaces, primarily due to gravitational forces. This can result in excessive deformation during the repair process, rendering the formation of proper repair impractical and leading to potential failure. In light of this rationale, conventional repair techniques are impractical for extensively damaged surfaces. Thus, this article proposes a novel repair methodology that is tailored to address large-area damage. Moreover, and departing from conventional practices involving the addition and subsequent subtraction of materials for precision machining, the proposed process endeavors to achieve more precise repair outcomes in a single operation. This paper introduces an innovative repair approach employing fused deposition modeling (FDM) to address the complexities associated with the repair of damaged polymer material parts. To mitigate geometric errors in the repaired structural components, beams with minimal deformation are printed using a compensation method. These beams then serve as supports for overlay printing. The paper outlines a methodology by which to determine the distribution of these supporting beams based on the shape of the damaged surface. A beam deformation model is established, and the printing trajectory of the compensated beam is calculated according to this model. Using the deformation model, the calculated deformation trajectories exhibit excellent fitting with the experimentally collected data, with an average difference between the two of less than 0.3 mm, validating the accuracy of the suspended beam deformation model. Based on the statistical findings, the maximum average deformation of the uncompensated sample is approximately 5.20 mm, whereas the maximum deformation of the sampled point after compensation measures around 0.15 mm. Consequently, the maximum deformation of the printed sample post-compensation is mitigated to roughly 3% of its pre-compensation magnitude. The proposed method in this paper was applied to the repair experiment of damaged curved surface components. A comparison was made between the point cloud data of the repaired surface and the ideal model of the component, with the average distance between them serving as the repair error metric. The mean distance between the point clouds of the repaired parts using the proposed repair strategy is 0.197 mm and the intact model surface is noticeably less than the mean distance corresponding to direct repair, at 0.830 mm. The repair error with compensatory support beams was found to be 76% lower than that without compensatory support beams. The surface without compensatory support beams exhibited gaps, while the surface with compensatory support beams appeared dense and complete. Experimental results demonstrate the effectiveness of the proposed method in significantly reducing the geometric errors in the repaired structural parts. The outcomes of the FDM repair method are validated through these experiments, affirming its practical efficacy. It is noteworthy that, although only PLA material was used in this study, the proposed method is general and effective for other polymer materials. This holds the potential to significantly reduce costs for the remanufacturing of widely used polymers.

A novel approach for residential load identification based on dynamic time warping

This paper proposes a new approach for load identification using a combination of fast Fourier transform (FFT) and dynamic time warping (DTW) techniques. In this approach, first, the current signals of each load device are measured individually and in various combinations to generate reference signals. FFT is then used to obtain meaningful features from the measured signals, while DTW is employed to classify the loads. This is the first known study to use the DTW algorithm for residential load identification in this structure. The proposed approach is constructed in two ways: supervised (DTW-SUP) and semi-supervised (DTW-SEM). In DTW-SEM, the reference signals for the load combinations are generated by summing the sample measurements of the individual loads, while in DTW-SUP, the actual sample measurements are used. The approach has been tested on six loads, including different load combinations. It has achieved high accuracy levels of over 90% in identifying single loads and load combinations using the error metrics Precision, Recall, and F-score. The proposed approach can potentially facilitate the implementation of energy-saving strategies in residential buildings and improve energy efficiency.

Real-time visual SLAM based YOLO-Fastest for dynamic scenes

Within the realm of autonomous robotic navigation, simultaneous localization and mapping (SLAM) serves as a critical perception technology, drawing heightened attention in contemporary research. The traditional SLAM systems perform well in static environments, but in the real physical world, dynamic objects can destroy the static geometric constraints of the SLAM system, further limiting its practical application in the real world. In this paper, a robust dynamic RGB-D SLAM system is proposed to expand the number of static points in the scene by combining with YOLO-Fastest to ensure the effectiveness of the geometric constraints model construction, and then based on that, a new thresholding model is designed to differentiate the dynamic features in the objection bounding box, which takes advantage of the double polyline constraints and the residuals after reprojection to filter the dynamic feature points. In addition, two Gaussian models are constructed to segment the moving objects in the bounding box in the depth image to achieve the effect similar to the instance segmentation under the premise of ensuring the computational speed. In this paper, experiments are conducted on dynamic sequences provided by the TUM dataset to evaluate the performance of the proposed method, and the results show that the root mean squared error metric of the absolute trajectory error of the algorithm of this paper has at least 80% improvement compared to ORB-SLAM2. Higher robustness in dynamic environments with both high and low dynamic sequences compared to DS-SLAM and Dynaslam, and can effectively provide intelligent localization and navigation for mobile robots.

BERT-Based Arabic Diacritization: A state-of-the-art approach for improving text accuracy and pronunciation

In order to accurately represent the meaning and pronunciation of Arabic words and sentences, the presence of diacritics plays a crucial role. Over the years, researchers have dedicated significant efforts to enhancing automated diacritization systems. This paper introduces a novel approach for Arabic diacritization utilizing Bidirectional Encoder representations from Transformers (BERT) models. To evaluate the effectiveness of the proposed approach, two publicly available datasets, namely the Arabic Diacritization (AD) dataset and the Tashkeela Processed (TP) dataset, were employed. The performance of the models was assessed using various error metrics, including Diacritic Error Rate (DER) and Word Error Rate (WER). The findings demonstrate the superior performance of BERT in the diacritization process, surpassing all models employed in other diacritization systems. On the AD dataset, the proposed system achieved state-of-the-art (SOTA) syntactic DER and WER of 1.14% and 3.34%, respectively. For morphological diacritization, the best results yielded a DER of 0.92% and a WER of 1.91%. These outcomes reflect a remarkable relative error reduction of over 30% compared to previous research. Additionally, on the TP dataset, the BERT models exhibited a substantial decrease in DER, reducing the benchmark from 4.0% to 1.11%. Furthermore, this study introduces a real-time diacritization system called SUKOUN, which offers diacritized text through a user-friendly website. A comparison with existing automatic diacritization tools, using six example texts, reveals the superior prediction accuracy and preservation of input format provided by SUKOUN.

Enhancing flood event predictions: Multi-objective calibration using gauge and satellite data

Rainfall-runoff models are widely used for predicting watershed runoff. The accuracy of the predictions heavily relies on effective model calibration, which is majorly influenced by the employed objective functions that measure the agreement between predictions and real-world observations. In this study, we propose a novel multi-objective calibration approach with the aim of improving the flood event predictions of a conceptual rainfall-runoff model applied to a catchment in the Austrian Alps. To achieve this, we include a total of five carefully selected objective functions into calibration. Three of these objective functions can be regarded as more classical and are roughly based on comparing absolute deviations, square of the residuals, as well as bias, correlation and variability of the flow. Another objective function relies on remote observations of the snow covered area, for which the MODIS Terra (MOD10A1.061) and Aqua (MYD10A1.061) datasets are used. The final objective function measures the differences in peak magnitude and timing for the largest events in the observation period. To showcase the proposed approach’s effectiveness, we compare the calibration and validation results with those obtained by a model that was only calibrated for two runoff objective functions and to carefully selected benchmark cases. The presented study reveals that our approach enhances the model’s capacity to depict snow cover dynamics, a crucial factor in accurately predicting snow-driven flooding events. Additionally, the results underscore that incorporating an error metric targeting magnitude and timing errors of large flood events can enhance the model’s ability to predict peak runoff. The results of this study further suggest that in general a more robust model can be obtained by incorporating additional error metrics that specifically target physical states and individual parts of the runoff curve into model calibration.

Deep Spectral Meshes: Multi-Frequency Facial Mesh Processing with Graph Neural Networks

With the rising popularity of virtual worlds, the importance of data-driven parametric models of 3D meshes has grown rapidly. Numerous applications, such as computer vision, procedural generation, and mesh editing, vastly rely on these models. However, current approaches do not allow for independent editing of deformations at different frequency levels. They also do not benefit from representing deformations at different frequencies with dedicated representations, which would better expose their properties and improve the generated meshes’ geometric and perceptual quality. In this work, spectral meshes are introduced as a method to decompose mesh deformations into low-frequency and high-frequency deformations. These features of low- and high-frequency deformations are used for representation learning with graph convolutional networks. A parametric model for 3D facial mesh synthesis is built upon the proposed framework, exposing user parameters that control disentangled high- and low-frequency deformations. Independent control of deformations at different frequencies and generation of plausible synthetic examples are mutually exclusive objectives. A Conditioning Factor is introduced to leverage these objectives. Our model takes further advantage of spectral partitioning by representing different frequency levels with disparate, more suitable representations. Low frequencies are represented with standardised Euclidean coordinates, and high frequencies with a normalised deformation representation (DR). This paper investigates applications of our proposed approach in mesh reconstruction, mesh interpolation, and multi-frequency editing. It is demonstrated that our method improves the overall quality of generated meshes on most datasets when considering both the L1 norm and perceptual Dihedral Angle Mesh Error (DAME) metrics.

Three-Dimensional Multi-Modality Registration for Orthopaedics and Cardiovascular Settings: State-of-the-Art and Clinical Applications.

The multimodal and multidomain registration of medical images have gained increasing recognition in clinical practice as a powerful tool for fusing and leveraging useful information from different imaging techniques and in different medical fields such as cardiology and orthopedics. Image registration could be a challenging process, and it strongly depends on the correct tuning of registration parameters. In this paper, the robustness and accuracy of a landmarks-based approach have been presented for five cardiac multimodal image datasets. The study is based on 3D Slicer software and it is focused on the registration of a computed tomography (CT) and 3D ultrasound time-series of post-operative mitral valve repair. The accuracy of the method, as a function of the number of landmarks used, was performed by analysing root mean square error (RMSE) and fiducial registration error (FRE) metrics. The validation of the number of landmarks resulted in an optimal number of 10 landmarks. The mean RMSE and FRE values were 5.26 ± 3.17 and 2.98 ± 1.68 mm, respectively, showing comparable performances with respect to the literature. The developed registration process was also tested on a CT orthopaedic dataset to assess the possibility of reconstructing the damaged jaw portion for a pre-operative planning setting. Overall, the proposed work shows how 3D Slicer and registration by landmarks can provide a useful environment for multimodal/unimodal registration.

Instantaneous Electricity Peak Load Forecasting Using Optimization and Machine Learning

Accurate instantaneous electricity peak load prediction is crucial for efficient capacity planning and cost-effective electricity network establishment. This paper aims to enhance the accuracy of instantaneous peak load forecasting by employing models incorporating various optimization and machine learning (ML) methods. This study examines the impact of independent inputs on peak load estimation through various combinations and subsets using multilinear regression (MLR) equations. This research utilizes input data from 1980 to 2020, including import and export data, population, and gross domestic product (GDP), to forecast the instantaneous electricity peak load as the output value. The effectiveness of these techniques is evaluated based on error metrics, including mean absolute error (MAE), mean square error (MSE), mean absolute percentage error (MAPE), root mean square error (RMSE), and R2. The comparison extends to popular optimization methods, such as particle swarm optimization (PSO), and the newest method in the field, including dandelion optimizer (DO) and gold rush optimizer (GRO). This comparison is made against conventional machine learning methods, such as support vector regression (SVR) and artificial neural network (ANN), in terms of their prediction accuracy. The findings indicate that the ANN and GRO approaches produce the least statistical errors. Furthermore, the correlation matrix indicates a robust positive linear correlation between GDP and instantaneous peak load. The proposed model demonstrates strong predictive capabilities for estimating peak load, with ANN and GRO performing exceptionally well compared to other methods.

Cross2SynNet: cross-device-cross-modal synthesis of routine brain MRI sequences from CT with brain lesion.

CT and MR are often needed to determine the location and extent of brain lesions collectively to improve diagnosis. However, patients with acute brain diseases cannot complete the MRI examination within a short time. The aim of the study is to devise a cross-device and cross-modal medical image synthesis (MIS) method Cross2SynNet for synthesizing routine brain MRI sequences of T1WI, T2WI, FLAIR, and DWI from CT with stroke and brain tumors. For the retrospective study, the participants covered four different diseases of cerebral ischemic stroke (CIS-cohort), cerebral hemorrhage (CH-cohort), meningioma (M-cohort), glioma (G-cohort). The MIS model Cross2SynNet was established on the basic architecture of conditional generative adversarial network (CGAN), of which, the fully convolutional Transformer (FCT) module was adopted into generator to capture the short- and long-range dependencies between healthy and pathological tissues, and the edge loss function was to minimize the difference in gradient magnitude between synthetic image and ground truth. Three metrics of mean square error (MSE), peak signal-to-noise ratio (PSNR), and structure similarity index measure (SSIM) were used for evaluation. A total of 230 participants (mean patient age, 59.77years ± 13.63 [standard deviation]; 163 men [71%] and 67 women [29%]) were included, including CIS-cohort (95 participants between Dec 2019 and Feb 2022), CH-cohort (69 participants between Jan 2020 and Dec 2021), M-cohort (40 participants between Sep 2018 and Dec 2021), and G-cohort (26 participants between Sep 2019 and Dec 2021). The Cross2SynNet achieved averaged values of MSE = 0.008, PSNR = 21.728, and SSIM = 0.758 when synthesizing MRIs from CT, outperforming the CycleGAN, pix2pix, RegGAN, Pix2PixHD, and ResViT. The Cross2SynNet could synthesize the brain lesion on pseudo DWI even if the CT image did not exhibit clear signal in the acute ischemic stroke patients. Cross2SynNet could achieve routine brain MRI synthesis of T1WI, T2WI, FLAIR, and DWI from CT with promising performance given the brain lesion of stroke and brain tumor.

An Autonomous Distributed Coordination Strategy for Sustainable Consumption in a Microgrid Based on a Bio-Inspired Approach

Distributed energy resources have demonstrated their potential to mitigate the limitations of large, centralized generation systems. This is achieved through the geographical distribution of generation sources that capitalize on the potential of their respective environments to satisfy local demand. In a microgrid, the control problem is inherently distributed, rendering traditional control techniques inefficient due to the impracticality of central governance. Instead, coordination among its components is essential. The challenge involves enabling these components to operate under optimal conditions, such as charging batteries with surplus solar energy or deactivating controllable loads when market prices rise. Consequently, there is a pressing need for innovative distributed strategies like emergent control. Inspired by phenomena such as the environmentally responsive behavior of ants, emergent control involves decentralized coordination schemes. This paper introduces an emergent control strategy for microgrids, grounded in the response threshold model, to establish an autonomous distributed control approach. The results, utilizing our methodology, demonstrate seamless coordination among the diverse components of a microgrid. For instance, system resilience is evident in scenarios where, upon the failure of certain components, others commence operation. Moreover, in dynamic conditions, such as varying weather and economic factors, the microgrid adeptly adapts to meet demand fluctuations. Our emergent control scheme enhances response times, performance, and on/off delay times. In various test scenarios, Integrated Absolute Error (IAE) metrics of approximately 0.01% were achieved, indicating a negligible difference between supplied and demanded energy. Furthermore, our approach prioritizes the utilization of renewable sources, increasing their usage from 59.7% to 86.1%. This shift not only reduces reliance on the public grid but also leads to significant energy cost savings.

Hybrid Uncertainty Calibration for Multimodal Sentiment Analysis

In open environments, multimodal sentiment analysis (MSA) often suffers from low-quality data and can be disrupted by noise, inherent defects, and outliers. In some cases, unreasonable multimodal fusion methods can perform worse than unimodal methods. Another challenge of MSA is effectively enabling the model to provide accurate prediction when it is confident and to indicate high uncertainty when its prediction is likely to be inaccurate. In this paper, we propose an uncertain-aware late fusion based on hybrid uncertainty calibration (ULF-HUC). Firstly, we conduct in-depth research on the issue of sentiment polarity distribution in MSA datasets, establishing a foundation for an uncertain-aware late fusion method, which facilitates organic fusion of modalities. Then, we propose a hybrid uncertainty calibration method based on evidential deep learning (EDL) that balances accuracy and uncertainty, supporting the reduction of uncertainty in each modality of the model. Finally, we add two common types of noise to validate the effectiveness of our proposed method. We evaluate our model on three publicly available MSA datasets (MVSA-Single, MVSA-Multiple, and MVSA-Single-Small). Our method outperforms state-of-the-art approaches in terms of accuracy, weighted F1 score, and expected uncertainty calibration error (UCE) metrics, proving the effectiveness of the proposed method.

Learning a Tracking Controller for Rolling μbots.

Micron-scale robots (μbots) have recently shown great promise for emerging medical applications. Accurate control of μbots, while critical to their successful deployment, is challenging. In this work, we consider the problem of tracking a reference trajectory using a μbot in the presence of disturbances and uncertainty. The disturbances primarily come from Brownian motion and other environmental phenomena, while the uncertainty originates from errors in the model parameters. We model the μbot as an uncertain unicycle that is controlled by a global magnetic field. To compensate for disturbances and uncertainties, we develop a nonlinear mismatch controller. We define the model mismatch error as the difference between our model's predicted velocity and the actual velocity of the μbot. We employ a Gaussian Process to learn the model mismatch error as a function of the applied control input. Then we use a least-squares minimization to select a control action that minimizes the difference between the actual velocity of the μbot and a reference velocity. We demonstrate the online performance of our joint learning and control algorithm in simulation, where our approach accurately learns the model mismatch and improves tracking performance. We also validate our approach in an experiment and show that certain error metrics are reduced by up to 40%.

A multi-glimpse deep learning architecture to estimate socioeconomic census metrics in the context of extreme scope variance

Convolutional Neural Networks (CNNs) are leveraged for a wide range of satellite imagery information extraction tasks. However, for tasks which seek to estimate aggregated information across highly variable geographic extents, existing techniques are subject to critical limitations. We engage with a specific case study exploring this challenge: estimating census variables across 2358 Mexican municipalities, which range in scope from 2.21 km2 (˜74,000 30 m pixels) to 72,417.9 km2 (millions of pixels). Building on recent literature which has illustrated the capability of deep learning to extract socioeconomic information from satellite imagery, we specifically seek to establish baseline metrics of error that might be expected when estimating a range of census variables based on coarse-resolution (Landsat) satellite imagery alone. For each of 52 variables, we implement a multi-glimpse recurrent attention model, in which we parametrically determine subsets of each municipality to sample across iterative steps. Results of a five-fold validation indicate that nearly half of the tested variables (22) can be estimated with r2 values greater than 0.75. Results suggest considerable promise for the use of satellite imagery to estimate socioeconomic factors in both historic time periods for which surveys were not conducted, as well as contemporary inaccessible regions.

Wind speed and wind power forecasting models

Sustainable energy resources have proved to be the best alternative in the wake of environmental degradation, depletion of ozone layer and ever-increasing demand for energy. Though wind energy is a promising resource, the non-linear nature and non-stationary characteristics of wind have remained a formidable challenge. Variability in wind power has posed numerous challenges in managing the power systems, especially in grid evacuation, penetration and integration. Forecasting wind is one of the powerful solutions to solve this problem. As the penetration of renewable energy sources is poised to increase in future, an accurate prediction can go a long way in helping the electricity grid to perform well. This article presents a review of existing research and recent trends in the forecasting of wind power and speed with a critical analysis of the contribution of every researcher. A review of forecasting technologies, data, time horizons, various forecasting approaches and error metrics has been presented in detail. The plethora of research issues that continue to challenge power system operators, wind farm owners and other stakeholders has been highlighted. The development of models for wind power or wind speed forecasting with excellent reliability and outstanding accuracy is the need of the hour.

Fuzzy C-Means clustering for physical model calibration and 7-day, 10-year low flow estimation in ungaged basins: comparisons to traditional, statistical estimates

In the northeast U.S., resource managers commonly apply 7-day, 10-year (7Q10) low flow estimates for protecting aquatic species in streams. In this paper, the efficacy of process-based hydrologic models is evaluated for estimating 7Q10s compared to the United States Geological Survey's (USGS) widely applied web-application StreamStats, which uses traditional statistical regression equations for estimating extreme flows. To generate the process-based estimates, the USGS's National Hydrologic Modeling (NHM-PRMS) framework (which relies on traditional rainfall-runoff modeling) is applied with 36 years of forcings from the Daymet climate dataset to a representative sample of ninety-four unimpaired gages in the Northeast and Mid-Atlantic U.S. The rainfall-runoff models are calibrated to the measured streamflow at each gage using the recommended NHM-PRMS calibration procedure and evaluated using Kling-Gupta Efficiency (KGE) for daily streamflow estimation. To evaluate the 7Q10 estimates made by the rainfall-runoff models compared to StreamStats, a multitude of error metrics are applied, including median relative bias (cfs/cfs), Root Mean Square Error (RMSE) (cfs), Relative RMSE (RRMSE) (cfs/cfs), and Unit-Area RMSE (UA-RMSE) (cfs/mi2). The calibrated rainfall-runoff models display both improved daily streamflow estimation (median KGE improving from 0.30 to 0.52) and 7Q10 estimation (smaller median relative bias, RMSE, RRMSE, and UA-RMSE, especially for basins larger than 100 mi2). The success of calibration is extended to ungaged locations using the machine learning algorithm Fuzzy C-Means (FCM) clustering, finding that traditional K-Means clustering (FCM clustering with no fuzzification factor) is the preferred method for model regionalization based on (1) Silhouette Analysis, (2) daily streamflow KGE, and (3) 7Q10 error metrics. The optimal rainfall-runoff models created with clustering show improvement for daily streamflow estimation (a median KGE of 0.48, only slightly below that of the calibrated models at 0.52); however, these models display similar error metrics for 7Q10 estimation compared to the uncalibrated models, neither of which provide improved error compared to the statistical estimates. Results suggest that the rainfall-runoff models calibrated to measured streamflow data provide the best 7Q10 estimation in terms of all error metrics except median relative bias, but for all models applicable to ungaged locations, the statistical estimates from StreamStats display the lowest error metrics in every category.

Predicting future sales of vegetable category based on grid search optimized long and short term neural networks

In this study, a method of automatic prediction of vegetable sales volume was proposed by using long and short term neural network (LSTM) to solve the problem of quantity judgment in vegetable supply chain management. This text constructed a trained and optimized LSTM forecasting model by collecting historical sales data and selling and wholesale prices of vegetable markets. After the optimal parameters were determined by grid search, the model underwent 10 training cycles. The model evaluation for each vegetable category led to the following conclusion: the model performed well in predicting sales. By using the evaluation metrics of Mean Square Error (MSE), Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE), it was found that the model predictions were close to the true values. In addition, the prediction results are summarized, showing the model's accurate prediction of vegetable sales volume. The results of the comparison chart between the real and predicted sales volume of each category of vegetables show that the model has good performance and accuracy in predicting the sales volume of vegetables, and the accurate forecast quantity can provide supply chain managers with more accurate inventory replenishment and pricing strategies.

GPROF V7 and beyond: assessment of current and potential future versions of the GPROF passive microwave precipitation retrievals against ground radar measurements over the continental US and the Pacific Ocean

Abstract. The Goddard Profiling Algorithm (GPROF) is used operationally for the retrieval of surface precipitation and hydrometeor profiles from the passive microwave (PMW) observations of the Global Precipitation Measurement (GPM) mission. Recent updates have led to GPROF V7, which has entered operational use in May 2022. In parallel, development is underway to improve the retrieval by transitioning to a neural-network-based algorithm called GPROF-NN. This study validates retrievals of liquid precipitation over snow-free and non-mountainous surfaces from GPROF V7 and multiple configurations of GPROF-NN against ground-based radar measurements over the conterminous United States (CONUS) and the tropical Pacific. GPROF retrievals from the GPM Microwave Imager (GMI) are validated over several years, and their ability to reproduce regional precipitation characteristics and effective resolution is assessed. Moreover, the retrieval accuracy for several other sensors of the constellation is evaluated. The validation of GPROF V7 indicates that the retrieval produces reliable estimates of liquid precipitation over the CONUS. During all four assessed years, annual mean precipitation is within 8 % of gauge-corrected radar measurements. Although biases of up to 25 % are observed over sub-regions of the CONUS and the tropical Pacific, the retrieval reliably reproduces each region's diurnal and seasonal precipitation characteristics. The effective resolution of GPROF V7 is found to be 51 km over the CONUS and 18 km over the tropical Pacific. GPROF V7 also produces robust precipitation estimates for the other sensors of the GPM constellation. The evaluation further shows that the GPROF-NN retrievals have the potential to significantly improve the GPM PMW precipitation retrievals. GPROF-NN 1D, the most basic neural network implementation of GPROF, improves the mean-squared error, mean absolute error, correlation and symmetric mean absolute percentage error of instantaneous precipitation estimates by about 20 % for GPROF GMI while the effective resolution is improved to 31 km over land and 15 km over oceans. The two GPROF-NN retrievals that are based on convolutional neural networks can further improve the accuracy up to the level of the combined radar–radiometer retrievals from the GPM core observatory. However, these retrievals are found to overfit on the viewing geometry at the center of the swath, reducing their overall accuracy to that of GPROF-NN 1D. For the other sensors of the constellation, the GPROF-NN retrievals produce larger biases than GPROF V7 and only GPROF-NN 3D achieves consistent improvements compared to GPROF V7 in terms of the other assessed error metrics. This points to shortcomings in the hydrometeor profiles or radiative transfer simulations used to generate the training data for the other sensors of the GPM constellation as a critical limitation for improving GPM PMW retrievals.

A general dual-pathway network for EEG denoising.

Scalp electroencephalogram (EEG) analysis and interpretation are crucial for tracking and analyzing brain activity. The collected scalp EEG signals, however, are weak and frequently tainted with various sorts of artifacts. The models based on deep learning provide comparable performance with that of traditional techniques. However, current deep learning networks applied to scalp EEG noise reduction are large in scale and suffer from overfitting. Here, we propose a dual-pathway autoencoder modeling framework named DPAE for scalp EEG signal denoising and demonstrate the superiority of the model on multi-layer perceptron (MLP), convolutional neural network (CNN) and recurrent neural network (RNN), respectively. We validate the denoising performance on benchmark scalp EEG artifact datasets. The experimental results show that our model architecture not only significantly reduces the computational effort but also outperforms existing deep learning denoising algorithms in root relative mean square error (RRMSE)metrics, both in the time and frequency domains. The DPAE architecture does not require a priori knowledge of the noise distribution nor is it limited by the network layer structure, which is a general network model oriented toward blind source separation.

Combining Deep Learning and Physical Models: A Benchmark Study on All‐Sky Imager‐Based Solar Nowcasting Systems

Intermittent solar irradiance due to passing clouds poses challenges for integrating solar energy into existing infrastructure. By making use of intrahour nowcasts (very short‐term forecasts), changing conditions of solar irradiance can be anticipated. All‐sky imagers, capturing sky conditions at high spatial and temporal resolution, can be the basis of such nowcasting systems. Herein, a deep learning (DL) model for solar irradiance nowcasts based on the transformer architecture is presented. The model is trained end‐to‐end using sequences of sky images and irradiance measurements as input to generate point‐forecasts up to 20 min ahead. Further, the effect of integrating this model into a hybrid system, consisting of a physics‐based model and smart persistence, is examined. A comparison between the DL and two hybrid models (with and without the DL model) is conducted on a benchmark dataset. Forecast accuracy for deterministic point‐forecasts is analyzed under different conditions using standard error metrics like root‐mean‐square error and forecast skill. Furthermore, spatial and temporal aggregation effects are investigated. In addition, probabilistic nowcasts for each model are computed via a quantile approach. Overall, the DL model outperforms both hybrid models under the majority of conditions and aggregation effects.