The rapid growth in digital image sharing, driven by advancements in internet and communication technologies, has raised concerns about image integrity, especially in sensitive fields like healthcare. This paper presents a fully blind fragile watermarking technique for authenticating and self-recovering color and grayscale medical images. The approach applies the Discrete Wavelet Transform (DWT) to the cover image, and the resulting subbands are divided into 3×3 blocks. One authentication and four recovery watermarks are then embedded into the least significant bits (LSBs) of each block. During the extraction phase, if tampering is detected, the model accurately localizes the altered areas, and a three-level recovery process, including a new inpainting technique, is used to recover the original image. Results based on two publically available datasets demonstrate that this method delivers high-quality watermarked images, achieves optimal watermark extraction accuracy, and maintains high sensitivity to various attacks. Additionally, it provides precise tamper localization and delivers high-quality recovered images, even with tampering rates as high as 60%.

Cyber-attack detection is crucial for securing Industrial Internet of Things (IIoT) systems. This study introduces advanced deep learning methodologies to identify potential cyber-attacks effectively in IIoT devices. Three novel stacked deep learning architectures, namely the StackMean, StackMax, and StackRF algorithms. These architectures aggregate and enhance the results of individual deep learning models. Specifically, StackMean computes average predicted class probabilities, StackMax selects maximum predicted class probabilities for more aggressive predictions, and StackRF leverages a random forest to aggregate base models. Theoretical analysis suggests that the proposed stacked deep learning model can boost detection accuracy compared to standalone single deep learning models. Moreover, these stacked models offer increased robustness against adversarial attacks by reducing reliance on specific neural network structures. Additionally, the synthetic minority oversampling technique (SMOTE) algorithm is integrated to address class imbalance challenges in the training dataset. Performance validation is conducted using three publicly available datasets. The detection performance is evaluated using five statistical scores. The results consistently indicate the superiority of the proposed stacked deep learning models over existing techniques. The effectiveness of the SMOTE algorithm is demonstrated through its ability to expand decision regions and minimize false negative signals during attack predictions. In addition, a statistical test is employed to compare the accuracy of individual models with the stacked models, demonstrating that the stacked models exhibit improved accuracy. By combining cutting-edge stacked deep learning architectures with strategic data augmentation techniques, this research significantly contributes to the robustness of cyber-attack detection within IIoT systems.

Accurate wind speed (WS) prediction in the Red Sea is essential for enhancing maritime operations, climate analysis, and monitoring ecosystems. Due to the region's complex oceanic and atmospheric patterns, this work introduces new models based on Transformer architectures to improve WS forecasting. Transformers are employed for their strength in handling sequential data and capturing time dependencies. A stacked model called StackedTrans has been developed to boost performance further and integrate multiple Transformer layers. The model's effectiveness is tested with WS data collected from ten locations across the Red Sea and evaluated using five statistical metrics. The results indicate that the StackedTrans model consistently outperforms other methods, such as LSTM, BiLSTM, GRU, BIGRU, and single Transformer models. The StackedTrans architecture performed a notable R2 score of 99.96, demonstrating its high precision in WS prediction

Electric vehicles (EVs) offer low carbon emissions; however, drivers frequently encounter difficulties scheduling charging sessions and locating available Charging Stations (CSs). These stations not only need to be conveniently located but also tailored to meet individual charging preferences. Addressing these challenges is crucial for overcoming barriers to the widespread adoption and efficiency of electric mobility, consequently enhancing the overall user experience. This paper introduces an innovative two-stage framework that improves the accessibility of EVCSs by integrating Graph Neural Networks (GNNs) with optimization algorithms. A bipartite graph representing user-station interactions is constructed in the first stage, and a GNN is utilized to leverage this structure. The GNNs efficiently capture complex relational patterns within the graph, enabling the generation of personalized station recommendations. In the subsequent stage, an optimization algorithm is employed to strategically assign users to these recommended stations. This algorithm considers station availability and proximity factors, ensuring optimal user assignments. The proposed recommender framework’s effectiveness is verified using data collected from the Yonne department in France. Experimental evaluations highlight the framework’s efficiency, achieving a high-performance measure of 98%, significantly reducing waiting times and maximizing user satisfaction for drivers compared to baseline approaches.

Electric vehicles (EVs) have long been recognized as a solution to the shortage of fossil fuels and the environmental problems associated with increasing CO2 emissions. However, charging an electric vehicle can take significant time at certain charging stations. Additionally, the limited deployment of charging stations is a significant barrier to the widespread adoption of electric mobility (e-mobility). In fact, many drivers struggle to locate a convenient charging station before their vehicle’s battery runs out. This study introduces a novel approach to addressing the issue of congestion at public charging stations and reducing the amount of time drivers spend waiting in line by predicting their occupancy. Previous research has relied on traditional Deep Learning (DL) techniques for prediction, which require centralized data collection. Nevertheless, each Charging Station Operator (CSO) holds sensitive data about its charging stations and users that cannot be shared with external parties. To address these privacy concerns, we propose a Federated Deep Learning approach where each CSO trains a DL model locally and then sends the model updates (or parameters) to a server for aggregation. Experiments on a real-world dataset demonstrate that predicting occupancy using the Federated Deep Learning approach achieves promising results (86,21% of accuracy and 91,49% of f1-score ), guarantees privacy, minimizes data transfer costs over the network, and allows individual CSOs to benefit from the rich datasets of others without sharing their sensitive data.

Summary Dust, or particulate matter (PM2.5), is among the most harmful pollutants negatively affecting human health. Predicting indoor PM2.5 concentrations is essential to achieve acceptable indoor air quality. This study aims to investigate data‐driven models to accurately predict PM 2.5 pollution. Notably, a comparative study has been conducted between twenty‐one machine learning and deep learning models to predict PM2.5 levels. Specifically, we investigate the performance of machine learning and deep learning models to predict ambient PM2.5 concentrations based on other ambient pollutants, including SO, NO, O, CO, and PM10. Here, we applied Bayesian optimization to optimally tune hyperparameters of the Gaussian process regression with different kernels and ensemble learning models (i.e., boosted trees and bagged trees) and investigated their prediction performance. Furthermore, to further enhance the forecasting performance of the investigated models, dynamic information has been incorporated by introducing lagged measurements in the construction of the considered models. Results show a significant improvement in the prediction performance when considering dynamic information from past data. Moreover, three methods, namely, random forest (RF), decision tree, and extreme gradient boosting, are applied to assess variables contribution and revealed that lagged PM2.5 data contribute significantly to the prediction performance and enables the construction of parsimonious models. Hourly concentration levels of ambient air pollution from the air quality monitoring network located in Seoul are employed to verify the prediction effectiveness of the studied models. Six measurements of effectiveness are used for assessing the prediction quality. Results showed that deep learning models are more efficient than the other investigated machine learning models (i.e., SVR, GPR, bagged and boosted trees, RF, and XGBoost). Also, the results showed that the bidirectional long short term memory (BiLSTM) and bidirectional gated recurrent units (BiGRU) networks produce higher performance than the investigated machine learning models (i.e., SVR, GPR, bagged and boosted trees, RF, and XGBoost) and deep learning models (i.e., LSTM, GRU, and convolutional neural network).

Summary Precisely detecting land cover changes aids in improving the analysis of the dynamics of the landscape and plays an essential role in mitigating the effects of desertification. Mainly, sensing desertification is challenging due to the high correlation between desertification and like‐desertification events (e.g., deforestation). An efficient and flexible deep learning approach is introduced to address desertification detection through Landsat imagery. Essentially, a generative adversarial network (GAN)‐based desertification detector is designed and for uncovering the pixels influenced by land cover changes. In this study, the adopted features have been derived from multi‐temporal images and incorporate multispectral information without considering image segmentation preprocessing. Furthermore, to address desertification detection challenges, the GAN‐based detector is constructed based on desertification‐free features and then employed to identify atypical events associated with desertification changes. The GAN‐detection algorithm flexibly learns relevant information from linear and nonlinear processes without prior assumption on data distribution and significantly enhances the detection's accuracy. The GAN‐based desertification detector's performance has been assessed via multi‐temporal Landsat optical images from the arid area nearby Biskra in Algeria. This region is selected in this work because desertification phenomena heavily impact it. Compared to some state‐of‐the‐art methods, including deep Boltzmann machine (DBM), deep belief network (DBN), convolutional neural network (CNN), as well as two ensemble models, namely, random forests and AdaBoost, the proposed GAN‐based detector offers superior discrimination performance of deserted regions. Results show the promising potential of the proposed GAN‐based method for the analysis and detection of desertification changes. Results also revealed that the GAN‐driven desertification detection approach outperforms the state‐of‐the‐art methods.

Accurately forecasting solar irradiance is indispensable in optimally managing and designing photovoltaic systems. It enables the efficient integration of photovoltaic systems in the smart grid. This paper introduces an innovative deep attention-driven model for solar irradiance forecasting. Notably, an extended version of the variational autoencoder (VAE) is introduced by amalgamating the desirable characteristics of the bidirectional LSTM (BiLSTM) and attention mechanism with the VAE model. Specifically, the introduced approach enables the conventional VAE’s ability to model temporal dependencies by incorporating BiLSTM at the VAE’s encoder side to better extract and learn temporal dependencies embed on the solar irradiance concentration measurements. In addition, the self-attention mechanism is embedded in the VAE’s encoder side following the BiLSTM to highlight pertinent features. The performance of the proposed model is evaluated through comparisons with the recurrent neural network (RNN), gated recurrent unit (GRU), LSTM, and BiLSTM. Measurements of solar irradiance in the US and Turkey are used to evaluate the investigated models. Results confirm the superior performance of the proposed model for solar irradiance forecasting over the other models (i.e., RNN, GRU, LSTM, and BiLSTM).

The demand for solar energy has rapidly increased throughout the world in recent years. However, anomalies in photovoltaic (PV) plants can reduce performances and result in serious consequences. Developing reliable statistical approaches able to detect anomalies in PV plants is vital to improving the management of these plants. Here, we present a statistical approach for detecting anomalies in the DC part of PV plants and partial shading. Firstly, we model the monitored PV plant. Then, we employ a generalized likelihood ratio test, which is a powerful anomaly detection tool, to check the residuals from the model and reveal anomalies in the supervised PV array. The proposed strategy is illustrated via actual measurements from a 9.54 PV plant.

Accurate detection of changes in land cover leads to better understanding of the dynamics of landscapes. This letter reports the development of a reliable approach to detecting changes in land cover based on remote sensing and radiometric data. This approach integrates the multivariate exponentially weighted moving average (MEWMA) chart with support vector machines (SVMs) for accurate and reliable detection of changes to land cover. Here, we utilize the MEWMA scheme to identify features corresponding to changed regions. Unfortunately, MEWMA schemes cannot discriminate between real changes and false changes. If a change is detected by the MEWMA algorithm, then we execute the SVM algorithm that is based on features corresponding to detected pixels to identify the type of change. We assess the effectiveness of this approach by using the remote-sensing change detection database and the SZTAKI AirChange benchmark data set. Our results show the capacity of our approach to detect changes to land cover.

This paper presents a computer vision-based methodology for human action recognition. First, the shape based pose features are constructed based on area ratios to identify the human silhouette in images. The proposed features are invariance to translation and scaling. Once the human body features are extracted from videos, different human actions are learned individually on the training frames of each class. Then, we apply the Adaboost algorithm for the classification process. We assessed the proposed approach using the UR Fall Detection dataset. In this study six classes of activities are considered namely: walking, standing, bending, lying, squatting, and sitting. Results demonstrate the efficiency of the proposed methodology.

Fault detection is important for effective and safe process operation. Partial least squares (PLS) has been used successfully in fault detection for multivariate processes with highly correlated variables. However, the conventional PLS-based detection metrics, such as the Hotelling's T2and the Q statistics are not well suited to detect small faults because they only use information about the process in the most recent observation. Exponentially weighed moving average (EWMA), however, has been shown to be more sensitive to small shifts in the mean of process variables. In this paper, a PLS-based EWMA fault detection method is proposed for monitoring processes represented by PLS models. The performance of the proposed method is compared with that of the traditional PLS-based fault detection method through a simulated example involving various fault scenarios that could be encountered in real processes. The simulation results clearly show the effectiveness of the proposed method over the conventional PLS method.

Fault detection is important for safe operation of various modern engineering systems. Partial least square (PLS) has been widely used in monitoring highly correlated process variables. Conventional PLS-based methods, nevertheless, often fail to detect incipient faults. In this paper, we develop new PLS-based monitoring chart, combining PLS with multivariate memory control chart, the multivariate exponentially weighted moving average (MEWMA) monitoring chart. The MEWMA are sensitive to incipient faults in the process mean, which significantly improves the performance of PLS methods and widen their applicability in practice. Using simulated distillation column data, we demonstrate that the proposed PLS-based MEWMA control chart is more effective in detecting incipient fault in the mean of the multivariate process variables, and outperform the conventional PLS-based monitoring charts.

The paper concerns the detection of fall events based on human silhouette shape variations. The detection of fall events is addressed from the statistical point of view as an anomaly detection problem. Specifically, the paper investigates the multivariate exponentially weighted moving average (MEWMA) control chart to detect fall events. Towards this end, a set of ratios for five partial occupancy areas of the human body for each frame are collected and used as the input data to MEWMA chart. The MEWMA fall detection scheme has been successfully applied to two publicly available fall detection databases, the UR fall detection dataset (URFD) and the fall detection dataset (FDD). The monitoring strategy developed was able to provide early alert mechanisms in the event of fall situations.

Fault detection is essential for safe operation of various engineering systems. Principal component analysis (PCA) has been widely used in monitoring highly correlated process variables. Conventional PCA-based methods, nevertheless, often fail to detect small or incipient faults. In this paper, we develop new PCA-based monitoring charts, combining PCA with multivariate memory control charts, such as the multivariate cumulative sum (MCUSUM) and multivariate exponentially weighted moving average (MEWMA) monitoring schemes. The multivariate control charts with memory are sensitive to small and moderate faults in the process mean, which significantly improves the performance of PCA methods and widen their applicability in practice. Using simulated data, we demonstrate that the proposed PCA-based MEWMA and MCUSUM control charts are more effective in detecting small shifts in the mean of the multivariate process variables, and outperform the conventional PCA-based monitoring charts.

This paper presents a statistical technique for detecting signs of abnormal situation generated by the influx of patients at emergency department (ED). The monitoring strategy developed was able to provide early alert mechanisms in the event of abnormal situations caused by abnormal patient arrivals to the ED. More specifically, This work proposed the application of autoregressive moving average (ARMA) models combined with the generalized likelihood ratio (GLR) test for anomaly-detection. ARMA was used as the modelling framework of the ARMA-based GLR anomaly-detection methodology. The GLR test was applied to the uncorrelated residuals obtained from the ARMA model to detect anomalies when the data did not fit the reference ARMA model. The ARMA-based GLR hypothesis testing scheme was successfully applied to the practical data collected from the database of the pediatric emergency department (PED) at Lille regional hospital center, France.

Monitoring charts play an important role in statistical quality control. Shewhart charts are among the most commonly used charts in process monitoring, and have seen many extensions for improved performance. Unfortunately, measured practical data are usually contaminated with noise, which degrade the detection abilities of the conventional Shewhart chart by increasing the rate of false alarms. Therefore, the effect of noise needs to be suppressed for enhanced process monitoring. Wavelet-based multiscale representation of data, which is a powerful feature extraction tool, has shown good abilities to efficiently separate deterministic and stochastic features. In this paper, the advantages of multiscale representation are exploited to enhance the fault detection performance of the conventional Shewhart chart by developing an integrated multiscale Shewhart algorithm. The performance of the developed algorithm is illustrated using two examples, one using synthetic data, and the other using simulated distillation column data. The simulation results clearly show the effectiveness of the proposed method over the conventional Shewhart chart and the conventional Shewhart chart applied on multiscale pre-filtered data.

One of the most popular multivariate statistical methods used for data-based process monitoring is Principal Component Analysis (PCA). In the absence of a process model, PCA has been successfully used as a data-based FD technique for highly correlated process variables. Some of the PCA detection indices include the T2 or Q statistics, which have their advantages and disadvantages. When a process model is available, however, the generalized likelihood ratio (GLR) test, which is a statistical hypothesis testing method, has shown good fault detection abili ties. In this work, a PCA-based GLR fault detection algorithm is developed to exploit the advantages of the GLR test in the absence of a process model. In fact, PCA is used to provide a modeling framework for the develop fault detection algorithm. The PCA-based GLR fault detection algorithm provides optimal properties by maximizing the detection probability of faults for a given false alarm rate. However, the presence of measurement noise and modeling errors increase the rate of false alarms. Therefore, to further improve the quality of fault detection, multiscale filtering is utilized to filter the residuals obtained from the PCA model, which helps suppress the effect on errors, and thus decrease the false alarm rate. The proposed fault detection methodology is demonstrated through its application to monitor the ozone level in the Upper Normandy region, France, and it is shown to effectively reduce the rate of false alarms whilst retaining the capability of detecting process faults.

Ozone is one of the lost serious air pollution problems. Monitoring abnormal changes in the concentration of ozone in the troposphere is of great interest because of its negative influence on human health, vegetation, and materials. Modeling ozone is very challenging because of the complexity of the ozone formation mechanisms in the troposphere and the uncertainty about the meteorological conditions in urban areas. In the absence of a process model, principal component analysis (PCA), which is a multivariate statistical technique, has been successfully used as a data-based fault detection (FD) method for highly correlated process variables. When a process model is available, however, the generalized likelihood ratio (GLR) test, which is a statistical hypothesis testing method, has shown good fault detection abilities. In this work, a PCA-based GLR fault detection algorithm is developed to exploit the advantages of the GLR test in the absence of a process model. In fact, PCA is used to provide a modeling framework for the develop fault detection algorithm. The developed PCA-based GLR FD algorithm is utilized to enhance monitoring the ozone concentrations in Upper Normandy, France. The performances of PCA and PCA-based GLR test are compared through two practical case studies, one involving a sensor fault and the other involving tropospheric ozone pollution in multiple measuring stations. The results show that the PCA-based GLR test can detect abnormal ozone levels with a smaller number of false alarms than the conventional PCA method.

Anomaly detection is addressed within a statistical framework. Often the statistical model is composed of two types of parameters : the informative parameters and the nuisance ones. The nuisance parameters are of no interest for detection but they are necessary to complete the model. In the case of unknown, non-random nuisance parameters, their elimination is unavoidable. Some approaches addressing the cases where the nuisance parameters, belonging to a subspace, interfere with the informative ones in a linear manner, use the theory of invariance to reject the nuisance. Sometimes this leads to a certain degradation of the detector performances because some faults become undetectable, masked by the nuisance. Nevertheless, in many cases the physical nature of nuisance parameters is (partially) known, and this knowledge may allow us to define inequality bounds to limit the variations of these parameters. The goal of this paper is to study the statistical performances of the constrained generalized likelihood ratio test used to detect an additive anomaly in the case of bounded nuisance parameters. An example of the integrity monitoring of GNSS train navigation illustrates the relevance of the proposed method.