Average MAE Research Articles

Personalized Ambient Pollution Estimation Based on Stationary-Camera-Taken Images Under Cross-Camera Information Sharing in Smart City

Timely and high-density air quality monitoring is essential for the development of future smart cities. The images captured from widely deployed stationary cameras can be transferred quickly via the Internet of Things (IoT) to facilitate ambient pollution estimation anytime anywhere. Imaged-based air pollution estimation is normally formulated as a supervised learning problem, relying on an extended number of image samples. However, individual stationary cameras can offer only very limited samples and scenes, whilst locally trained estimation models can easily overfit. A global method was proposed to address this challenge. The global model was trained via images captured from different cameras. However, such a model is less effective in extracting local features from scenes. A personalized method is therefore proposed to improve not only the generalization of the estimation model but also to preserve the local characteristics of individual cameras. Our personalized method consists of a two-stage architecture: (1) images from different cameras are used to train the global estimation model to avoid overfitting due to fixed scenes and small sample size; (2) the global model is further refined by images captured from individual cameras separately for adapting local characteristics. To evaluate our proposed personalized method, a large dataset was constructed, based on stationary camera-taken images captured in Hong Kong, consisting of different pollution measurements, including PM2.5, PM10, NO2 and O3. As compared to the local model, our proposed personalized model has reduced average MAE by 5.68% and average SMAPE by 6.82%, and improved average r by 4.69%.

Data-driven prediction for curved pipe jacking performance during underwater excavation of ancient shipwreck using an attention-based graph convolutional network approach

Thrust is a crucial parameter to ensure the quality of underwater curved pipe jacking construction. A prediction framework based on hybrid deep learning is proposed to realize real-time prediction of thrust. The method contains graph convolutional neural network (GCN), gated recurrent unit (GRU), and self-attention (SA) mechanism. To enhance the interpretability of the method, Shapley Additive Explanations (SHAP), and causal discovery models are introduced within the framework. Taking the salvage project of the “Yangtze Estuary No. 2″ shipwreck as a case study, the prediction performance of the model is verified. The results show that the proposed model achieves excellent performance in thrust prediction for underwater curved pipe jacking, with an average MAE of 11.07 kN, RMSE of 13.96 kN, and R2 value of 0.934. The SHAP analysis reveals that angle, buried depth, and stratum changes exert significant influences on thrust. Furthermore, the spatial correlation and temporal correlation among features are revealed. Compared to other models, the proposed model exhibits superior prediction accuracy and is capable of handling various time step sizes, thereby providing valuable decision support for curved pipe jacking thrust prediction.

Dual-Path Information Fusion and Twin Attention-Driven Global Modeling for Solar Irradiance Prediction.

Accurate prediction of solar irradiance holds significant value for renewable energy usage and power grid management. However, traditional forecasting methods often overlook the time dependence of solar irradiance sequences and the varying importance of different influencing factors. To address this issue, this study proposes a dual-path information fusion and twin attention-driven solar irradiance forecasting model. The proposed framework comprises three components: a residual attention temporal convolution block (RACB), a dual-path information fusion module (DIFM), and a twin self-attention module (TSAM). These components collectively enhance the performance of multi-step solar irradiance forecasting. First, the RACB is designed to enable the network to adaptively learn important features while suppressing irrelevant ones. Second, the DIFM is implemented to reinforce the model's robustness against input data variations and integrate multi-scale features. Lastly, the TSAM is introduced to extract long-term temporal dependencies from the sequence and facilitate multi-step prediction. In the solar irradiance forecasting experiments, the proposed model is compared with six benchmark models across four datasets. In the one-step predictions, the average performance metrics RMSE, MAE, and MAPE of the four datasets decreased within the ranges of 0.463-2.390 W/m2, 0.439-2.005 W/m2, and 1.3-9.2%, respectively. Additionally, the average R2 value across the four datasets increased by 0.008 to 0.059. The experimental results indicate that the model proposed in this study exhibits enhanced accuracy and robustness in predictive performance, making it a reliable alternative for solar irradiance forecasting.

Multi-step-ahead and interval carbon price forecasting using transformer-based hybrid model.

Accurate and stable carbon price forecasts serve as a reference for assessing the stability of the carbon market and play a vital role in enhancing investment and operational decisions. However, realizing this goal is still a significant challenge, and researchers usually ignore multi-step-ahead and interval forecasting due to the non-linear and non-stationary characteristics of carbon price series and its complex fluctuation features. In this study, a novel hybrid model for accurately predicting carbon prices is proposed. The proposed model combines multi-step-ahead and interval carbon price forecasting based on the Hampel identifier (HI), time-varying filtering-based empirical mode decomposition (TVFEMD), and transformer model. First, HI identifies and corrects outliers in carbon price. Second, TVFEMD decomposes carbon price into several intrinsic mode functions (imfs) to reduce the non-linear and non-stationarity of carbon price to obtain more regular features in series. Next, these imfs are reconstructed by sample entropy (SE). Subsequently, the orthogonal array tuning method is used to optimize the transformer model's hyperparameters to obtain the optimal model structure. Finally, after hyperparameter optimization and quantile loss function, the transformer is used to perform multi-step-ahead and interval forecasting on each part of the reconstruction, and the final prediction result is obtained by summing them up. Five pilot carbon trading markets in China were selected as experimental objects to verify the proposed model's prediction performance. Various benchmark models and evaluation indicators were selected for comparison and analysis. Experimental results show that the proposed HI-TVFEMD-transformer hybrid model achieves an average MAE of 0.6546, 1.3992, 1.6287, and 2.2601 for one-step, three-step, five-step, and ten-step-ahead forecasting, respectively, which significantly outperforms other models. Furthermore, interval forecasts almost always have a PICI above 0.95 at a confidence interval of 0.1, thereby indicating the effectiveness of the hybrid model in describing the uncertainty in the forecasts. Therefore, the proposed hybrid model is a reliable carbon price forecasting tool that can provide a dependable reference for policymakers and investors.

An Improved VMD-LSTM Model for Time-Varying GNSS Time Series Prediction with Temporally Correlated Noise

GNSS time series prediction plays a significant role in monitoring crustal plate motion, landslide detection, and the maintenance of the global coordinate framework. Long short-term memory (LSTM) is a deep learning model that has been widely applied in the field of high-precision time series prediction and is often combined with Variational Mode Decomposition (VMD) to form the VMD-LSTM hybrid model. To further improve the prediction accuracy of the VMD-LSTM model, this paper proposes a dual variational modal decomposition long short-term memory (DVMD-LSTM) model to effectively handle noise in GNSS time series prediction. This model extracts fluctuation features from the residual terms obtained after VMD decomposition to reduce the prediction errors associated with residual terms in the VMD-LSTM model. Daily E, N, and U coordinate data recorded at multiple GNSS stations between 2000 and 2022 were used to validate the performance of the proposed DVMD-LSTM model. The experimental results demonstrate that, compared to the VMD-LSTM model, the DVMD-LSTM model achieves significant improvements in prediction performance across all measurement stations. The average RMSE is reduced by 9.86% and the average MAE is reduced by 9.44%; moreover, the average R2 increased by 17.97%. Furthermore, the average accuracy of the optimal noise model for the predicted results is improved by 36.50%, and the average velocity accuracy of the predicted results is enhanced by 33.02%. These findings collectively attest to the superior predictive capabilities of the DVMD-LSTM model, thereby demonstrating the reliability of the predicted results.

Fast simulation and prediction of urban pluvial floods using a deep convolutional neural network model

Urban pluvial floods induced by rainstorms can cause severe losses to human lives and property. Fast and accurate simulation and prediction of urban pluvial flood are of significance for disaster prevention and mitigation. However, physics-based models still experience excessive computational time when used for flood simulation and prediction. In this study, we explore a deep learning (DL) approach employing convolutional neural networks (CNN) as an alternative model to achieve fast prediction of urban floods. We first cluster a rainstorm-inundation database generated using a physics-based coupled model, then we develop a CNN model for predicting the spatial and temporal evolution of rainstorm inundation, and finally we compare the effectiveness of the CNN model with that of the coupled model and three other classical machine learning (ML) models. The results show that: 1) The inundation water depths predicted using the CNN model are close to those predicted using the coupled model, and the average PCC, MAE and RMSE metrics under a test rainstorm reach 0.983, 0.020 m and 0.086 m, respectively. 2) The CNN model reproduces well the trend of water depth in each model grid cell over time, especially for a heavily inundated grid. 3) The predicted effectiveness of the CNN model outperforms an extreme gradient boosting (XGBoost) model, followed by a multi-objective random forest (MORF) model and K-nearest neighbor (KNN) model. 4) The computational speed of the CNN model is extremely fast. The model can simulate inundation water depth with a spatial resolution of 8 m by 8 m (about 74 km2) and a temporal resolution of 30 min for a 6-hour lead time within 12 s, which is 600 times faster than that of a coupled model. We confirm that the CNN model with clustering method is a powerful surrogate model for fast simulation of urban pluvial flood, providing an important reference for the use of DL in early warning and mitigation in relation to urban flood disasters.

Adaptive multi-source data fusion vessel trajectory prediction model for intelligent maritime traffic

Multi-source vessel automatic identification system (AIS) data gathered in the maritime Internet of Things (IoT) system has diverse data characteristics, such as sparse satellite data and terrestrial receiver’s limited coverage. However, it is challenging for the maritime IoT system to provide efficient and accurate intelligent trajectory prediction services while applying multi-source AIS data. Therefore, improving the adaptability and accuracy of vessel trajectory prediction methodologies adapted to varied AIS datasets is essential to enhancing marine traffic safety and transportation efficiency. This study proposes an adaptive data fusion (ADF) model based on multi-source AIS data for predicting vessel trajectories in maritime traffic. The same marine mobile service identifier (MMSI) and timestamp fuse the multi-source AIS data. Deep learning methods are employed to perform feature learning and increase adaptability. This study also utilizes adjacency distance to enhance the correlation between dataset trajectory trajectories. Finally, we validate the benefits of the ADF model using actual terrestrial and satellite AIS data along the west coast of the United States (US). Regarding average RMSE and MAE, the ADF model reduces the prediction error by 46.51% and 52.7% when fusing multi-source historical trajectories compared to current studies. Moreover, multiple ablation experiments have proved our model’s superiority.

Attention‐based model for dynamic IR drop prediction with multi‐view features

AbstractDynamic IR drop prediction based on machine learning has been studied in recent years. However, most proposed models used all input features extracted from circuits or manually selected parts of raw features as inputs, which failed to differentiate the order of priority among input features in a flexible manner. In this paper, QuantumForest to vector‐based dynamic IR drop prediction is introduced. With the sparse attention mechanism brought by QuantumForest, important attributes of circuits are weighed more heavily than others. A new multi‐view feature creation method is also proposed and a novel regional distance feature is built up subsequently. The performance is evaluated on two chip designs with real simulation vectors. The experiment results indicate that the prediction result of the method outperforms other prominent methods for dealing with machine learning based IR drop analysis, reaching an average MAE of only 1.457 on two designs.

Day-ahead photovoltaic power forecasting using hybrid K-Means++ and improved deep neural network

The relationship between environmental factors with the power output of photovoltaic (PV) stations is unclear due to the non-linear characteristics of PV systems, which is challenging for PV power forecasting technology. To cope with these challenges, a hybrid forecasting approach called hybrid K-Means++ and Deep Neural Network with input and output adjusting structures (K-IAOA-DNN) is proposed to accurately predict PV power output. The proposed forecasting approach designs several features for K-Means++ to search for similar samples, reducing the complexity of the following forecasting model. Due to changeable environmental conditions and characteristics of PV systems like degradation, the prediction result of a forecasting model may deviate from the expected one. Therefore, IAOA-DNN model is developed by using a DNN adopting the two proposed structures e.g., the input adjusting structure and the output adjusting structure. The input adjusting structure uses several trainable parameters to determine the importance of each meteorological input for further feature extracting by DNN. The output structure in the prediction model is used for analyzing features extracted from DNN to generate a scalar factor fine-tuning the output of DNN. Additionally, the proposed method provides more accurate forecasting results with average RMSE, NRMSE, and MAE values of 14.43, 0.048, and 9.53 kW, respectively.

Spatial prediction and influencing factors identification of potential toxic element contamination in soil of different karst landform regions using integration model

The prediction of contamination distribution of potentially toxic elements (PTEs) in soils of Guangxi province, China and the identification of their controlling factors pose great challenges due to diverse bedrock types, intense leaching and weathering, and discontinuous terrain distributions. Herein, we integrated the random forest (RF) and empirical Bayesian kriging (EBK) to interpret and predict complex PTEs contamination distribution from three different karst landform regions (fenglin, fengcong, isolated peak plain) in Guangxi province. The modeling results are compared with the commonly used ordinary kriging and regression-kriging. In this study, our developed RF-EBK model combines the advantages of the RF and EBK model to promote the prediction accurately and efficiently. In this study, it was shown that the integration RF-EBK model exhibited desirable for Cd and As concentrations, with R2 of 0.89 and 0.83, respectively. The average RMSE and MAE of integration RF-EBK model decreased by 39% and 44%, respectively, relative to the regression-kriging with the second highest accuracy. Furthermore, the modeling results showed that approximately 41.96% and 18.96% of total area was classified as Cd and As polluted and above regions (Igeo >0) in Guangxi province, respectively. Higher Cd concentration was observed in the soil of fenglin and fengcong regions than that in isolated peak plain region due to the secondary enrichment and parent rock inheritance, while the As concentration exhibited no significant difference among the three regions. The modeling results indicated that the elevated Cd concentration might be associated with soil CaO concentration and alkaline soil environment, whereas As concentration tended to be increased with the elevating Fe2O3 concentrations in weakly acidic soil environment. This result confirmed the applicability and effectiveness of integration model in predicting complex spatial patterns of soil PTEs and identifying their controlling factors.

Long-Term Glucose Forecasting for Open-Source Automated Insulin Delivery Systems: A Machine Learning Study with Real-World Variability Analysis

Glucose forecasting serves as a backbone for several healthcare applications, including real-time insulin dosing in people with diabetes and physical activity optimization. This paper presents a study on the use of machine learning (ML) and deep learning (DL) methods for predicting glucose variability (GV) in individuals with open-source automated insulin delivery systems (AID). A three-stage experimental framework is employed in this work to systematically implement and evaluate ML/DL methods on a large-scale diabetes dataset collected from individuals with open-source AID. The first stage involves data collection, the second stage involves data preparation and exploratory analysis, and the third stage involves developing, fine-tuning, and evaluating ML/DL models. The performance and resource costs of the models are evaluated alongside relative and proportional errors for 17 GV metrics. Evaluation of fine-tuned ML/DL models shows considerable accuracy in glucose forecasting and variability analysis up to 48 h in advance. The average MAE ranges from 2.50 mg/dL for long short-term memory models (LSTM) to 4.94 mg/dL for autoregressive integrated moving average (ARIMA) models, and the RMSE ranges from 3.7 mg/dL for LSTM to 7.67 mg/dL for ARIMA. Model execution time is proportional to the amount of data used for training, with long short-term memory models having the lowest execution time but the highest memory consumption compared to other models. This work successfully incorporates the use of appropriate programming frameworks, concurrency-enhancing tools, and resource and storage cost estimators to encourage the sustainable use of ML/DL in real-world AID systems.

Research on Fast CFD Simulation of Automobile Flow Field Based on Artificial Intelligence

Fast automotive aerodynamic evaluation can extremely reduce the design cycles of automotive. This paper establishes a rapid simulation model of the automobile flow field based on the artificial intelligence surrogate model. The end-to-end network is used to map the geometric model, incoming flow conditions, and result. The decoder is used to splice high-dimensional and low-dimensional features to achieve feature sharing. The average MAE error of the optimal model is 5.249%. The average calculation time of a single example is 1.2968s, which is about 0.62% of CFD solver. The simulation result demonstrate that the deep learning method can not only accelerate the calculation, but also can improve the design efficiency of automobile aerodynamic profile with high accuracy.

Ordinal Regression Based on the Distributional Distance for Tabular Data

Ordinal regression is used to classify instances by considering ordinal relation between labels. Existing methods tend to decrease the accuracy when they adhere to the preservation of the ordinal relation. Therefore, we propose a distributional knowledge-based network (DK-net) that considers ordinal relation while maintaining high accuracy. DK-net focuses on image datasets. However, in industrial applications, one can find not only image data but also tabular data. In this study, we propose DK-neural oblivious decision ensemble (NODE), an improved version of DK-net for tabular data. DK-NODE uses NODE for feature extraction. In addition, we propose a method for adjusting the parameter that controls the degree of compliance with the ordinal relation. We experimented with three datasets: WineQuality, Abalone, and Eucalyptus dataset. The experiments showed that the proposed method achieved high accuracy and small MAE on three datasets. Notably, the proposed method had the smallest average MAE on all datasets.

Two-step fusion method for generating 1 km seamless multi-layer soil moisture with high accuracy in the Qinghai-Tibet plateau

Current remote sensing techniques fail to observe and generate large scale multi-layer soil moisture (SM) due to the inherent features of the satellite sensors. The lack of comprehensive understanding of multi-layer SM hinders the sustainable development of agriculture, hydrology, and food security. In order to overcome the depth barrier of traditional SM assimilation and downscaling methods, we developed a Two-step Multi-layer SM Downscaling (TMSMD) framework by fusing multi-source remotely sensed, reanalysis, and in-situ data through both machine learning and state-of-the-art deep learning models to generate multi-layer SM. The produced multi-layer SM was characterized by high resolution (1 km), high spatio-temporal continuity (cloud-free and daily), and high accuracy (i.e., 3H data). Firstly, the coarse resolution SMAP SM was downscaled to 1 km spatial resolution using LightGBM to weaken the effects of scale mismatch issue and provide high-resolution input for the subsequent calibration. Results indicated that the downscaled SMAP SM remained high consistency with the original SMAP SM product. With the high-resolution inputs, we calibrated the downscaled SMAP SM using multi-layer in-situ SM through state-of-the-art attention-based LSTM. Results demonstrated that the average PCC, RMSE, ubRMSE, and MAE were improved by 22.3 %, 50.7 %, 26.2 %, and 56.7 % compared to SMAP L4 SM while 38.5 %, 52.1 %, 29.5 %, and 58.7 % compared to downscaled SMAP SM. Further spatio-temporal and comparative analysis confirmed that the multi-layer SM produced by the TMSMD framework had excellent performance in capturing the spatial and temporal dynamics. In conclude, the proposed TMSMD framework successfully generated 3H multi-layer SM data and is promising for accurate assessment and monitoring in agriculture, water resources, and environmental domains.

Hourly stepwise forecasting for solar irradiance using integrated hybrid models CNN-LSTM-MLP combined with error correction and VMD

Accurate and reliable solar irradiance forecasting is critical for distribution planning and modern smart grid management and dispatch. However, due to the time series of solar irradiance with the nonlinearity and nonstationarity, some researches up to now are still unsatisfactory in terms of prediction accuracy and model generalization ability. Therefore, to improve the comprehensive performance of the model, a novel forecasting framework of hourly stepwise forecasting for solar irradiance using integrated hybrid models CNN-LSTM-MLP combined with error correction and variational mode decomposition (VMD) was proposed. In this paper, three different datasets were pre-processed and then the step-by-step predictions of seven different methods were completed based on the proposed procedures. Finally, comprehensive analysis results of the model indicated that the prediction scheme proposed in this paper made full use of, step-by-step prediction, VMD method, integrated hybrid model, and error correction four major advantages greatly improved the anti-interference ability of the model, so that the average performance metrics RMSE, nRMSE, MAE, and R2 of the three datasets reach to 12.53 W/m2, 6.04 %, 7.65 W/m2, and 99.79 %, respectively. Moreover, the optimal promoting percentages of the R2(PR2)indicator compared to the persistence model (Per) is by 20.17 %. It is found that the model is superior to a large number of traditional alternative approaches in terms of accuracy and robustness, which may provide a reference for comprehensive performance optimization of the model in the future.

Investigation of Hillslope Vineyard Soil Water Dynamics Using Field Measurements and Numerical Modeling

Soil heterogeneities can impact hillslope hydropedological processes (e.g., portioning between infiltration and runoff), creating a need for in-depth knowledge of processes governing water dynamics and redistribution. The presented study was conducted at the SUPREHILL Critical Zone Observatory (CZO) (hillslope vineyard) in 2021. A combination of field investigation (soil sampling and monitoring campaign) and numerical modeling with hydrological simulator HYDRUS-1D was used to explore the water dynamics in conjunction with data from a sensor network (soil water content (SWC) and soil-water potential (SWP) sensors), along the hillslope (hilltop, backslope, and footslope). Soil hydraulic properties (SHP) were estimated based on (i) pedotransfer functions (PTFs), (ii) undisturbed soil cores, and (iii) sensor network data, and tested in HYDRUS. Additionally, a model ensemble mean from HYDRUS simulations was calculated with PTFs. The highest agreement of simulated with observed SWC for 40 cm soil depth was found with the combination of laboratory and field data, with the lowest average MAE, RMSE and MAPE (0.02, 0.02, and 5.34%, respectively), and highest average R2 (0.93), while at 80 cm soil depth, PTF model ensemble performed better (MAE = 0.03, RMSE = 0.03, MAPE = 7.55%, R2 = 0.81) than other datasets. Field observations indicated that heterogeneity and spatial variability regarding soil parameters were present at the site. Over the hillslope, SWC acted in a heterogeneous manner, which was most pronounced during soil rewetting. Model results suggested that the incorporation of field data expands model performance and that the PTF model ensemble is a feasible option in the absence of laboratory data.

Hydrogen yield prediction for supercritical water gasification based on generative adversarial network data augmentation

Machine learning methods can accurately predict the hydrogen yield of supercritical water gasification (SCWG), which can provide a reference for SCWG optimization. However, the time cost of obtaining SCWG data by experiments is high, and the prediction accuracy of the SCWG model is low when the sample number is small. For this problem, a novel sample generation method based on the generative adversarial network (GAN) and the hybrid strategy is proposed to achieve data augmentation for SCWG hydrogen yield prediction. The hybrid strategy that combines the cross-validation and looping iteration methods can reduce the randomness of the GAN, thereby improving the generated sample quality. In addition, the least squares support vector regression is adopted to construct a hydrogen yield prediction model. A case study was conducted on a small sample SCWG dataset of Yimin lignite with reaction temperatures in the range of 650–850 °C. The results show that after expanding the small sample dataset by the single GAN, the average RMSE and MAE of the prediction model are reduced by 20.43% and 11.58%, respectively. The proposed sample generation method is applied to the SCWG hydrogen yield prediction. The average RMSE and MAE are as low as 1.1737 mol/kg and 1.0171 mol/kg, respectively, while the average R2 reaches 0.9457. The average RMSE and MAE are further reduced by 18.44% and 15.29% based on the single GAN model, respectively. It indicates that the hybrid strategy improves the quality and reliability of the generated samples, and the proposed method can effectively achieve data augmentation for SCWG hydrogen yield prediction, which can provide ideas for the small sample SCWG modeling.

Mapping of disease-specific Oxford Knee Score onto EQ-5D-5L utility index in knee osteoarthritis

BackgroundEQ5D is a generic measure of health. It provides a single index value for health status that can be used in the clinical and economic evaluation of healthcare. Oxford Knee Score (OKS) is a joint-specific outcome measure tool designed to assess symptoms and function in osteoarthritis patients after joint replacement surgery. Though widely used, it has the disadvantage of lacking health index value. To fill the gap between functional and generic questionnaires with economic value, we linked generic EQ-5D-5L to the specific OKS to give a single index value for health status in KOA patients.Questions/purposesDeveloping and evaluating an algorithm to estimate EuroQoL generic health utility scores (EQ-5D-5L) from the disease-specific OKS using data from patients with knee osteoarthritis (KO).Patients and methodsThis is a cross-sectional study of 571 patients with KO. We used four distinct mapping algorithms: Cumulative Probability for Ordinal Data, Penalized Ordinal Regression, CART (Classification and Regression Trees), and Ordinal random forest. We compared the resultant models’ degrees of accuracy.ResultsMobility was best predicted by penalized regression with pre-processed predictors, usual activities by random forest, pain/discomfort by cumulative probability with pre-processed predictors, self-care by random forest with RFE (recursive feature elimination) predictors, and anxiety/depression by CART with RFE predictors. Model accuracy was lowest with anxiety/depression and highest with mobility and usual activities. Using available country value sets, the average MAE was 0.098 ± 0.022, ranging from 0.063 to 0.142; and the average MSE was 0.020 ± 0.008 ranging from 0.008 to 0.042.ConclusionsThe current study derived accurate mapping techniques from OKS to the domains of EQ-5D-5L, allowing for the computation of QALYs in economic evaluations. A machine learning-based strategy offers a viable mapping alternative that merits further exploration.

Scale-aware CNN for crowd density estimation and crowd behavior analysis

Visual surveillance is a highly demanded system in various real-time applications for civil and military sectors. For huge gatherings such as rallies and stadiums, crowd monitoring becomes a tedious task. Moreover, crowd density estimation also plays an important role to improve the performance of crowd monitoring system. Several techniques have been presented for crowd counting or density estimation and crowd behavior analysis. The crowd scenes contain noise, occlusion, cluttered environment which creates complexity to analyze the crowd behavior. To overcome the challenges of crowd monitoring, we present a combined solution to perform the crowd density estimation and crowd behavior analysis. The crowd density estimation is carried out by developing a new CNN based architecture which considers scale information and overcome the issues of scale variations. In order to deal with this issue of scale variation, we introduce a scale-aware attention module where multiple branches of self-attention module are concatenated to improve the scale variation realization. Further, we present the crowd behavior analysis model which uses motion map and energy level distribution based features to detect the abnormal crowd behavior. The Proposed approach achieves average accuracy of abnormal event detection as 98.95, and crowd counting accuracy as 97.60. Furthermore, the average MAE and MSE is achieved as 58.60, 98.55 for ShanghaiTech Part A dataset and 7.55, 8.50 for Part B dataset, respectively. Similarly, the MAE and MSE for UCF 50 dataset is obtained as 210.20, and 260.80, respectively which shows a significant improvement over other techniques discussed in the experimental study.

A Neuromechanical Model-Based Strategy to Estimate the Operator's Payload in Industrial Lifting Tasks.

One of the main technological barriers hindering the development of active industrial exoskeleton is today represented by the lack of suitable payload estimation algorithms characterized by high accuracy and low calibration time. The knowledge of the payload enables exoskeletons to dynamically provide the required assistance to the user. This work proposes a payload estimation methodology based on personalized Electromyography-driven musculoskeletal models (pEMS) combined with a payload estimation method we called "delta torque" that allows the decoupling of payload dynamical properties from human dynamical properties. The contribution of this work lies in the conceptualization of such methodology and its validation considering human operators during industrial lifting tasks. With respect to existing solutions often based on machine learning, our methodology requires smaller training datasets and can better generalize across different payloads and tasks. The proposed payload estimation methodology has been validated on lifting tasks with 0kg, 5kg, 10kg and 15kg, resulting in an average MAE of about 1.4 Kg. Even if 5kg and 10Kg lifting tasks were out of the training set, the MAE related to these tasks are 1.6 kg and 1.1 kg, respectively, demonstrating the generalizing property of the proposed methodology. To the best of the authors' knowledge, this is the first time that an EMG-driven model-based approach is proposed for human payload estimation.