Search

With the advancement of information and communication technology, the integration of smart systems into discrete sequential processes has been realized in manufacturing systems. Specifically, to optimize the efficiency of operations and production, accurate and rapid detection of faults is crucial. Numerous studies have attempted to identify faults in various electromechanical systems; however, the performance and effectiveness often depend on the quality of the collected signals and the system operation methodology. For instance, the measurements of various types of analog sensor signals—such as voltage, current, 3-axis acceleration, and temperature—vary according to the installation locations and positions within a system. Moreover, although an automated manufacturing system typically operates under predefined control sequences, identifying common signal patterns for every healthy or faulty state is challenging. Consequently, we focused on detecting faults using only image data captured from on-site camera in automated sequential manufacturing processes, without relying on additional manufacturing information. We proposed a method based on convolutional long short-term memory (LSTM) networks for the next visual status prediction and an unsupervised statistical process control chart to classify the current status as normal or faulty. Following the detection of a fault, we conducted unsupervised fault type clustering to provide valuable insights for maintenance efforts. The validation and verification demonstrated accurate fault detection (94.2%) and effective fault type clustering in two distinct manufacturing processes of a USB packaging system. Since no information other than the original process images captured by the on-site camera were used for fault detection and fault type classification, it can be assumed that the proposed methodology can be easily applied to other or more complex manufacturing processes. [ABSTRACT FROM AUTHOR]

Introduction: The precise forecasting of Significant wave height(SWH) is vital to ensure the safety and efficiency of aquatic activities such as ocean engineering, shipping, and fishing. Methods: This paper proposes a deep learning model named SAC-ConvLSTM to perform 24-hour prediction with the SWH in the South China Sea. The long-term prediction capability of the model is enhanced by using the attention mechanism and context vectors. The prediction ability of the model is evaluated by mean absolute error (MAE), root mean square error (RMSE), mean square error (MSE), and Pearson correlation coefficient (PCC). Results: The experimental results show that the optimal input sequence length for the model is 12. Starting from 12 hours, the SAC-ConvLSTM model consistently outperforms other models in predictive performance. For the 24-hour prediction, this model achieves RMSE, MAE, and PCC values of 0.2117 m, 0.1083 m, and 0.9630, respectively. In addition, the introduction of wind can improve the accuracy of wave prediction. The SAC-ConvLSTM model also has good prediction performance compared to the ConvLSTM model during extreme weather, especially in coastal areas. Discussion: This paper presents a 24-hour prediction of SWH in the South China Sea. Through comparative validation, the SAC-ConvLSTM model outperforms other models. The inclusion of wind data enhances the model's predictive capability. This model also performs well under extreme weather conditions. In physical oceanography, variables related to SWH include not only wind but also other factors such as mean wave period and sea surface air pressure. In the future, additional variables can be incorporated to further improve the model's predictive performance. [ABSTRACT FROM AUTHOR]

While machine learning methods have been successful in predicting air pollution, current deep learning models usually focus only on the time-based connection of air quality monitoring stations or the complex link between PM2.5 levels and explanatory factors. Due to the lack of effective integration of spatial correlation, the prediction model shows poor performance in PM2.5 prediction tasks. Predicting air pollution levels accurately over a long period is difficult because of the changing levels of correlation between past pollution levels and the future. In order to address these challenges, the study introduces a Convolutional Long Short-Term Memory (ConvLSTM) network-based neural network model with multiple feature extraction for forecasting PM2.5 levels in air quality prediction. The technique is composed of three components. The model-building process of this article is as follows: Firstly, we create a complex network layout with multiple branches to capture various temporal features at different levels. Secondly, a convolutional module was introduced to enable the model to focus on identifying neighborhood units, extracting feature scales with high spatial correlation, and helping to improve the learning ability of ConvLSTM. Next, the module for spatiotemporal fusion prediction is utilized to make predictions of PM2.5 over time and space, generating fused prediction outcomes that combine characteristics from various scales. Comparative experiments were conducted. Experimental findings indicate that the proposed approach outperforms ConvLSTM in forecasting PM2.5 concentration for the following day, three days, and seven days, resulting in a lower root mean square error (RMSE). This approach excels in modeling spatiotemporal features and is well-suited for predicting PM2.5 levels in specific regions. [ABSTRACT FROM AUTHOR]

Abstract In this study, the deep learning algorithm of Convolutional Neural Network long short-term memory (CNN–LSTM) is used to classify various jewelry rocks such as agate, turquoise, calcites, and azure from various historical periods and styles related to Shahr-e Sokhteh. Here, the CNN–LSTM architecture includes utilizing CNN layers for the extraction of features from input data mixed with LSTMs for supporting sequence forecasting. It should be mentioned that interpretable deep learning-assisted laser induced breakdown spectroscopy helped achieve excellent performance. For the first time, this paper interprets the Convolutional LSTM effectiveness layer by layer in self-adaptively obtaining LIBS features and the quantitative data of major chemical elements in jewelry rocks. Moreover, Lasso method is applied on data as a factor for investigation of interoperability. The results demonstrated that LIBS can be essentially combined with a deep learning algorithm for the classification of different jewelry songs. The proposed methodology yielded high accuracy, confirming the effectiveness and suitability of the approach in the discrimination process.

Human behavior is regarded as one of the most complex notions present nowadays, due to the large magnitude of possibilities. These behaviors and actions can be distinguished as normal and abnormal. However, abnormal behavior is a vast spectrum, so in this work, abnormal behavior is regarded as human aggression or in another context when car accidents occur on the road. As this behavior can negatively affect the surrounding traffic participants, such as vehicles and other pedestrians, it is crucial to monitor such behavior. Given the current prevalent spread of cameras everywhere with different types, they can be used to classify and monitor such behavior. Accordingly, this work proposes a new optimized model based on a novel integrated wavelet-based channel augmentation unit for classifying human behavior in various scenes, having a total number of trainable parameters of 5.3 m with an average inference time of 0.09 s. The model has been trained and evaluated on four public datasets: Real Live Violence Situations (RLVS), Highway Incident Detection (HWID), Movie Fights, and Hockey Fights. The proposed technique achieved accuracies in the range of 92% to 99.5% across the used benchmark datasets. Comprehensive analysis and comparisons between different versions of the model and the state-of-the-art have been performed to confirm the model’s performance in terms of accuracy and efficiency. The proposed model has higher accuracy with an average of 4.97%, and higher efficiency by reducing the number of parameters by around 139.1 m compared to other models trained and tested on the same benchmark datasets.

Sea surface temperature (SST) prediction has received increasing attention in recent years due to its paramount importance in the various fields of oceanography. Existing studies have shown that neural networks are particularly effective in making accurate SST predictions by efficiently capturing spatiotemporal dependencies in SST data. Among various models, the ConvLSTM framework is notably prominent. This model skillfully combines convolutional neural networks (CNNs) with recurrent neural networks (RNNs), enabling it to simultaneously capture spatiotemporal dependencies within a single computational framework. To overcome the limitation that CNNs primarily capture local spatial information, in this paper we propose a novel model named DatLSTM that integrates a deformable attention transformer (DAT) module into the ConvLSTM framework, thereby enhancing its ability to process more complex spatial relationships effectively. Specifically, the DAT module adaptively focuses on salient features in space, while ConvLSTM further captures the temporal dependencies of spatial correlations in the SST data. In this way, DatLSTM can adaptively capture complex spatiotemporal dependencies between the preceding and current states within ConvLSTM. To evaluate the performance of the DatLSTM model, we conducted short-term SST forecasts in the Bohai Sea region with forecast lead times ranging from 1 to 10 days and compared its efficacy against several benchmark models, including ConvLSTM, PredRNN, TCTN, and SwinLSTM. Our experimental results show that the proposed model outperforms all of these models in terms of multiple evaluation metrics short-term SST prediction. The proposed model offers a new predictive learning method for improving the accuracy of spatiotemporal predictions in various domains, including meteorology, oceanography, and climate science.

Traffic flow prediction is a crucial research area in traffic management. Accurately predicting traffic flow in each area of the city over the long term can enable city managers to make informed decisions regarding the allocation of urban transportation resources in the future. The existing traffic flow prediction models either give insufficient attention to the interactions of long-lasting spatio-temporal regions or extract spatio-temporal features in a single scale, which ignores the identification of traffic flow patterns at various scales. In this paper, we present a multi-scale spatio-temporal information fusion model using non-local networks, which fuses traffic flow pattern features at multiple scales in space and time, complemented by non-local networks to construct the global direct dependence relationship between local areas and the entire region of the city in space and time in the past. The proposed model is evaluated through experiments and is shown to outperform existing benchmark models in terms of prediction performance. [ABSTRACT FROM AUTHOR]

Human behavior is regarded as one of the most complex notions present nowadays, due to the large magnitude of possibilities. These behaviors and actions can be distinguished as normal and abnormal. However, abnormal behavior is a vast spectrum, so in this work, abnormal behavior is regarded as human aggression or in another context when car accidents occur on the road. As this behavior can negatively affect the surrounding traffic participants, such as vehicles and other pedestrians, it is crucial to monitor such behavior. Given the current prevalent spread of cameras everywhere with different types, they can be used to classify and monitor such behavior. Accordingly, this work proposes a new optimized model based on a novel integrated wavelet-based channel augmentation unit for classifying human behavior in various scenes, having a total number of trainable parameters of 5.3 m with an average inference time of 0.09 s. The model has been trained and evaluated on four public datasets: Real Live Violence Situations (RLVS), Highway Incident Detection (HWID), Movie Fights, and Hockey Fights. The proposed technique achieved accuracies in the range of 92 % to 99.5 % across the used benchmark datasets. Comprehensive analysis and comparisons between different versions of the model and the state-of-the-art have been performed to confirm the model's performance in terms of accuracy and efficiency. The proposed model has higher accuracy with an average of 4.97 % , and higher efficiency by reducing the number of parameters by around 139.1 m compared to other models trained and tested on the same benchmark datasets. [ABSTRACT FROM AUTHOR]

Predicting volatility is important for asset predicting, option pricing and hedging strategies because it cannot be directly observed in the financial market. The dynamics of the volatility surface is difficult to estimate. In this paper, we establish a novel architecture based on physics-informed neural networks and convolutional transformers. The performance of the new architecture is directly compared to other well-known deep-learning architectures, such as standard physics-informed neural networks, convolutional long-short term memory (ConvLSTM), and self-attention ConvLSTM. Numerical evidence indicates that the proposed physics-informed convolutional transformer network achieves a superior performance than other methods. [ABSTRACT FROM AUTHOR]

In recent years, there are multiple temperature predictive models have demonstrated the effectiveness and outperformances of applying different deep neural architectures, such as convolutional neural network (CNN) and recurrent neural network (RNN) for the temperature forecasting task in forms of time-series analysis problem in comparing with previous traditional machine learning based techniques. However, up to this time, there are still several limitations of existing deep learning-based temperature predictive methods related to the capability of efficiently integrating extra information resources into the temperature data learning process. Moreover, the high-noised/chaotic fluctuations within daily temperature data also lead to downgrades in the accuracy performance of recent deep learning-based techniques. To overcome these challenges, in this paper, we proposed a novel integrated dual attention mechanism with the Convolutional Long Short-Term Memory Network (LSTM), called as: DAttConvLSTM. Our proposed DAttCovLSTM supports to effectively capture the chaotic and dynamic temporal information from daily temperature data, thus significantly improve the prediction results. Extensive experiments and comparative studies in real-world datasets demonstrated the effectiveness of our proposed model in comparing with recent state-of-the-art baselines. [ABSTRACT FROM AUTHOR]

While machine learning methods have been successful in predicting air pollution, current deep learning models usually focus only on the time-based connection of air quality monitoring stations or the complex link between PM2.5 levels and explanatory factors. Due to the lack of effective integration of spatial correlation, the prediction model shows poor performance in PM2.5 prediction tasks. Predicting air pollution levels accurately over a long period is difficult because of the changing levels of correlation between past pollution levels and the future. In order to address these challenges, the study introduces a Convolutional Long Short-Term Memory (ConvLSTM) network-based neural network model with multiple feature extraction for forecasting PM2.5 levels in air quality prediction. The technique is composed of three components. The model-building process of this article is as follows: Firstly, we create a complex network layout with multiple branches to capture various temporal features at different levels. Secondly, a convolutional module was introduced to enable the model to focus on identifying neighborhood units, extracting feature scales with high spatial correlation, and helping to improve the learning ability of ConvLSTM. Next, the module for spatiotemporal fusion prediction is utilized to make predictions of PM2.5 over time and space, generating fused prediction outcomes that combine characteristics from various scales. Comparative experiments were conducted. Experimental findings indicate that the proposed approach outperforms ConvLSTM in forecasting PM2.5 concentration for the following day, three days, and seven days, resulting in a lower root mean square error (RMSE). This approach excels in modeling spatiotemporal features and is well-suited for predicting PM2.5 levels in specific regions.

In terms of SST prediction tasks by machine learning approaches, high-resolution and accurate prediction can hardly be achieved with easily accessible SST data. This is because most of the SST data is spatiotemporally discrete and sparse, meaning a small sample size since it is hard to learn the spatiotemporal correlation given the data of limited density. This paper presents a numerical computation-based few-shot learning method for intelligent SST prediction. The proposed method generates synthetic SST sequences via ROMS and merges them with SST data captured by satellites to deal with small sample size. Convolutional LSTM (Conv-LSTM) network is trained end-to-end in order to learn spatiotemporal correlation of time-varying SST and eventually obtain high-resolution and accurate SST predictions. SST data from August to December in 2011 are employed for model training, and the prediction results are compared to reliable SST reanalysis datasets in the experiment, which shows fine performance of the proposed method. [ABSTRACT FROM AUTHOR]

Fine particulate matter (PM2.5) concentration in ambient air has become a major concern across the globe. All major cities of India have reported an elevated concentration of PM2.5 that has severe consequences to the health, economy, and ecosystem of the region. As a result, it becomes imperative to develop adequate tools for forecasting particulate matter concentration. Most of the research works mostly focused on single-step prediction horizon, thereby limiting their use. In the present work, a hybrid model has been proposed to forecast multi-step ahead concentrations of PM2.5 in ambient air across India covering different agroclimatic zones. The hybrid model architecture was an encoder-decoder-based sequence to sequence model framework that was built with convolutional long short-term memory (LSTM), bidirectional LSTM and 3D convolution neural network. The model was tested across 26 Indian cities covering 13 major agroclimatic zones of India. The performance of the model was also analysed for consecutive hour sequential prediction taking last 24-h data as input to the model. The model output was also compared with signal to noise ratio to explore the reason for variations in model performance. A distinct trend was found between signal to noise ratio and model output. As noise increases, the model performances suffer. Overall, the model was found to be stable as its performance errors across different time horizon has little variations. The proposed model has the potential to be used for long-term forecasting by incorporating other predictor variables series. [ABSTRACT FROM AUTHOR]

Recently, the traditional transmission line fault diagnosis approaches cannot handle the variables' dynamic coupling properties, and they also ignore the local structure feature information during the feature extraction. To figure out these issues, a novel enhanced feature extraction based convolutional LSTM (ECLSTM) approach is developed to diagnose the transmission line fault in this paper. Our work has three main contributions: (1) To tackle the dynamic coupling characteristics of the process variables, the statistics analysis (SA) method is first employed to calculate different statistical features of the transmission line's original data, where the original datasets are transformed into the subsequently used statistics datasets; (2) The statistics comprehensive feature preserving (SCFP) is then proposed to maintain both the global and local structure features of the constructed statistics datasets, where the locality structure preserving technique is incorporated into the principal component analysis (PCA) model to extract the features from the statistics datasets; (3) To effectively diagnose the transmission line's fault, the SCFP based convolutional LSTM fault diagnosis scheme is constructed to classify the global and local statistical structure features of fault snapshot dataset, because of its ability to exploit the temporal dependencies and spatial correlations of the extracted statistical features. Detailed experiments and comparisons on the datasets of the simulated power system are performed to prove the excellent performance of the ECLSTM based fault diagnosis scheme. [ABSTRACT FROM AUTHOR]

Demand prediction for on-demand food delivery (ODFD) is of great importance to the operation and transportation resource utilization of ODFD platforms. This paper addresses short-term ODFD demand prediction using an end-to-end deep learning architecture. The problem is formulated as a spatial–temporal prediction. The proposed model is composed of convolutional long short-term memory (ConvLSTM), and convolutional neural network (CNN) units with encoder–decoder structure. Specifically, long short-term memory (LSTM) networks are a type of recurrent neural network capable of learning order dependence in sequence prediction problems. The convolution unit is responsible for capturing spatial attributes, while the LSTM part is adopted to learn temporal attributes. Additionally, an attentional model is designed and integrated to enhance the prediction performance by addressing the spatial variation in demand. The proposed approach is compared to several baseline models using a historical ODFD dataset from Shenzhen, China. Results indicate that the proposed model obtains the highest prediction accuracy by capturing both spatial and temporal correlations with attention information focusing on different parts of the input series. [ABSTRACT FROM AUTHOR]

The South Platte river system contains a mixture of natural streams, reservoirs, and pipeline projects that redirect water to front range communities in Colorado. At many timepoints, a simple persistence model is the best predictor for flow from pipelines and reservoirs but at other times, flows change based on snowmelt and inputs such as reservoir fill rates, local weather, and anticipated demand. Here we find that a convolutional Long Short-Term Memory (LSTM) network is well suited to modeling flow in parts of this basin that are strongly impacted by water projects as well as ones that are relatively free from direct human modifications. Furthermore, it is found that including an active learning component in which separate Convolutional Neural Networks (CNNs) are used to classify and then select the data that is then used for training a convolutional LSTM network is advantageous. Models specific for each gauge are created by transfer of parameter from a base model and these gauge-specific models are then fine-tuned based a curated subset of training data. The result is accurate predictions for both natural flow and human influenced flow using only past river flow, reservoir capacity, and historical temperature data. In 14 of the 16 gauges modeled, the error in the prediction is reduced when using the combination of on-the-fly classification by CNN followed by analysis by either a persistence or convolutional LSTM model. The methods designed here could be applied broadly to other basins and to other situations where multiple models are needed to fit data at different times and locations.

The accurate detection and recognition of human actions play a pivotal role in aerial surveillance, enabling the identification of potential threats and suspicious behavior. Several approaches have been presented to address this problem, but the limitation still remains in devising an accurate and robust solution. To this end, this paper presents an effective action recognition framework for aerial surveillance, employing the YOLOv8-Pose keypoints extraction algorithm and a customized sequential ConvLSTM (Convolutional Long Short-Term Memory) model for classifying the action. We performed a detailed experimental evaluation and comparison on the publicly available Drone Action dataset. The evaluation and comparison of the proposed framework with several existing approaches on the publicly available Drone Action dataset demonstrate its effectiveness, achieving a very encouraging performance. The overall accuracy of the framework on three provided dataset splits is 74%, 80%, and 70%, with a mean accuracy of 74.67%. Indeed, the proposed system effectively captures the spatial and temporal dynamics of human actions, providing a robust solution for aerial action recognition. [ABSTRACT FROM AUTHOR]

Abstract Deep learning has achieved tremendous success in various applications owing to its robust feature representations of complex high-dimensional nonlinear data. Financial time-series prediction is no exception. Hence, the volatility trend prediction in financial time series (FTS) has been an active topic for several decades. Inspired by generative adversarial networks (GAN), which have been studied extensively in image processing and have achieved excellent results, we present the ordinal regression GAN for financial volatility trends (ORGAN-FVT) method for the end-to-end multi-classification task of FTS. An improved generative model based on convolutional long short-term memory (ConvLSTM) and multilayer perceptron (MLP) is proposed to capture temporal features effectively and mine the data distribution of volatility trends (short, neutral, and long) from given FTS data. Meanwhile, ordinal regression is leveraged for the discriminator to improve the multi-classification performance, making the model more practical. Finally, we empirically compare ORGAN-FVT with several state-of-the-art approaches on three real-world stock datasets: MICROSOFT(MSFT), Tesla(TSLA), and The People’s Insurance Company of China(PAICC). ORGAN-FVT demonstrated significantly better AUC and F1 scores, at most 20.81% higher than its competitors.

The human body's temporal fluctuation is referred to as human activity. The preservation of cultural heritage, the development of video recommendation systems, the support of learners via tutoring systems will benefit from the capture and evaluation of dance-related multimedia content. Because of its detailed hand gesture, Indian classical dance (ICD) classification is still an enthralling field of study. This provides a framework for analyzing various computer vision and deep learning ideas. Automated teaching solutions across all disciplines, from traditional to online forums, become unavoidable through changes in learning habits. ICD also becomes a crucial component of a thriving culture and heritage that must be updated and preserved at all costs. The dance involves complex positions like self-hands-occlusion and full-body rotation. The main objective of this study is, in the Bharatanatyam dancing style we proposed a framework for categorizing hasta mudras. Our Convolution Neural Network - Long Short Term Memory (CNN-LSTM) deep knowledge architecture for Indian Classical Dancing (ICD) categorization now includes a new hand posture signature. By guessing where people's hands would be, we rated dance performances. This architecture assesses hand poses using information and information pruning, while a dance instructor application assesses the time and accuracy of student dances. 252 YouTube videos of the Bharatanatyam dance form has been used to make up the dataset used in our research.This study offers a methodology with three-phase deep learning techniques. Then, using the pre-trained paradigm TensorFlow EfficientNet –UNet, which aids us in determining any hand position within the frame, we extracted the appropriate joint locations of the hands from each video frame. Then, cosine similarity was used to identify or correlate the indicated action factors. Finally, using key details from the hand pose, we categorized it and trained the Convolution Neural Network - Long Short Term Memory (CNN-LSTM) network structure using the classification system's training dataset. Regarding factors like accuracy, F1-score, AUC curve, recall and precision, the proposed CNN-LSTM structure for classifying hand mudras is compared with Convolutional LSTM Long Term Recurrent Convolutional Network(LRCN), Multilayer Perceptron (MLP), LSTM and 3D Convolutional Layer (CONV3D). As a result, it was found that throughout the examination process, the proposed CNN-LSTM classification structure achieved 98.53% accuracy, 99.04% precision, 98.49% recall, 99.12% AUC score, and 98.74% F1-score. This achieves 94.03% accuracy, 93.13% precision, 94.76% recall, 96.06% AUC score, and 93.53% F1-score during the training method. [ABSTRACT FROM AUTHOR]

Noise annoyance is recognized as an expression of physiological and psychological strain in acoustical environment. The studies on prediction of noise annoyance and parametric sensitivity analysis of factors affecting it have been rarely reported in India. A hybrid ConvLSTM technique was developed in the study to predict traffic-induced noise annoyance in 484 people based on ambient noise levels, as well as survey information. Ambient noise levels were obtained at different locations of Dhanbad city using sound level meter at varying intervals, viz. 09AM–12PM, 03PM–06PM, and 08PM–11PM. The proposed method was compared with some well-known neural network techniques such as K-nearest neighbors (KNN), artificial neural network (ANN), recurrent neural network (RNN), and long-short-term memory (LSTM). The experimental results indicate that the proposed method outperforms other techniques and can be a reliable approach for prediction of noise annoyance with an accuracy of 93.8%. It can be concluded from noise maps that the noise levels in all locations of the Dhanbad city were higher than 70 dB(A) and noise sensitivity is the most important input variable of traffic-induced noise annoyance, followed by honking noise, education, exposure hours, LAeq, sleeping disorder, and chronic disease. The study shall facilitate in developing a decision support tool for prediction of noise annoyance and promoting implementation of suitable public policy in urban cities. [ABSTRACT FROM AUTHOR]

Traffic flow prediction is a crucial research area in traffic management. Accurately predicting traffic flow in each area of the city over the long term can enable city managers to make informed decisions regarding the allocation of urban transportation resources in the future. The existing traffic flow prediction models either give insufficient attention to the interactions of long-lasting spatio-temporal regions or extract spatio-temporal features in a single scale, which ignores the identification of traffic flow patterns at various scales. In this paper, we present a multi-scale spatio-temporal information fusion model using non-local networks, which fuses traffic flow pattern features at multiple scales in space and time, complemented by non-local networks to construct the global direct dependence relationship between local areas and the entire region of the city in space and time in the past. The proposed model is evaluated through experiments and is shown to outperform existing benchmark models in terms of prediction performance.

We developed a Convolutional LSTM (ConvLSTM) network to predict shoulder joint reaction forces using 3D shoulder kinematics data containing 30 different shoulder activities from eight human subjects. We considered simulation outcomes from the AnyBody musculoskeletal model as the baseline force dataset to validate ConvLSTM model predictions. Results showed a good correlation (>80% accuracy, r ≥ 0.82) between ConvLSTM predicted and AnyBody estimated force values, the generalization of the developed model for novel task type (p-value = 0.07 ∼ 0.33), and a better prediction accuracy for the ConvLSTM model than conventional CNN and LSTM models. [ABSTRACT FROM AUTHOR]

Sea surface temperature (SST) prediction has received increasing attention in recent years due to its paramount importance in the various fields of oceanography. Existing studies have shown that neural networks are particularly effective in making accurate SST predictions by efficiently capturing spatiotemporal dependencies in SST data. Among various models, the ConvLSTM framework is notably prominent. This model skillfully combines convolutional neural networks (CNNs) with recurrent neural networks (RNNs), enabling it to simultaneously capture spatiotemporal dependencies within a single computational framework. To overcome the limitation that CNNs primarily capture local spatial information, in this paper we propose a novel model named DatLSTM that integrates a deformable attention transformer (DAT) module into the ConvLSTM framework, thereby enhancing its ability to process more complex spatial relationships effectively. Specifically, the DAT module adaptively focuses on salient features in space, while ConvLSTM further captures the temporal dependencies of spatial correlations in the SST data. In this way, DatLSTM can adaptively capture complex spatiotemporal dependencies between the preceding and current states within ConvLSTM. To evaluate the performance of the DatLSTM model, we conducted short-term SST forecasts in the Bohai Sea region with forecast lead times ranging from 1 to 10 days and compared its efficacy against several benchmark models, including ConvLSTM, PredRNN, TCTN, and SwinLSTM. Our experimental results show that the proposed model outperforms all of these models in terms of multiple evaluation metrics short-term SST prediction. The proposed model offers a new predictive learning method for improving the accuracy of spatiotemporal predictions in various domains, including meteorology, oceanography, and climate science. [ABSTRACT FROM AUTHOR]

With traffic network becoming increasingly complicated, traffic flow prediction has important practical significance for the management of traffic roads and public safety. For example, an accurate taxi demand prediction can help to improve efficiency of vehicle scheduling and reduce traffic congestion. The main issue of flow prediction is how to extract the information of complex spatio-temporal dependencies and interactions between arrival and departure. To solve these problems, we develop a deep learning method based on time-dependent attention convolutional LSTM (TDAConvLSTM) in which a time-dependent attention mechanism is designed to learn similarities of historical traffic flows among different time intervals and a fusion mechanism is introduced to aggregate the feature information produced by convolutional LSTM and attention module. And then, the result of the feature aggregation is fed to a multi-layer deconvolutional network to gain the results of flow prediction. Experimental studies on two real-life datasets indicate that TDAConvLSTM achieves better results than the compared models. The source code of our proposed method is available at the URL1. [ABSTRACT FROM AUTHOR]

Oil prices are inherently volatile, and they used to suffer from many fluctuations and changes. Therefore, oil prices prediction is the subject of many studies in the field, some researchers concentrated on the key factors that could influence the prediction accuracy, while the others focused on designing models that forecast the prices with high accuracy. To help the institutions and companies to hedge against any sudden changes and develop right decisions that support the global economy, in this project the concept of cascade networks model to predict the crude oil prices has been adopted, that can be considered relatively as new initiative in the field. The model is used to predict the Iraqi oil prices since as its commonly known that the economy in Iraq is totally depend on oil. Therefore, it is vital to develop a better perception about the crude oil price dynamics because its volatility can cause a sudden economic crisis. [ABSTRACT FROM AUTHOR]

Human brain is known to have a layered structure and perform not only feedforward process from lower layer to upper layer, but also feedback process from upper layer to lower layer. Neural network is a mathematical model of the function of neurons, and several models are proposed until now. Although neural network imitates the human brain, everyone uses only feedforward process and direct feedback process from upper layer to lower layer is not used in prediction process. Therefore, in this paper, we propose Feedback U-Net using convolutional LSTM. Our model is a segmentation model using convolutional LSTM and feedback process. The output of U-Net at the first round is fed back to the input, and our method re-considers the segmentation result at the second round. By using convolutional LSTM, the features are extracted well based on the features extracted at the first round. On both of the Drosophila cell image and Mouse cell image datasets, our model outperformed conventional U-Net which uses only feedforward process. [ABSTRACT FROM AUTHOR]

Accurate predictive modeling of traffic flow is critically important as it allows transportation users to make wise decisions to circumvent traffic congestion regions. The advanced development of sensing technology makes big data more affordable and accessible, meaning that data-driven methods have been increasingly adopted for traffic flow prediction. Although numerous data-driven methods have been introduced for traffic flow predictions, existing data-driven methods cannot consider the correlation of the extracted high-dimensional features and cannot use the most relevant part of the traffic flow data to make predictions. To address these issues, this work proposes a decoder convolutional LSTM network, where the convolutional operation is used to consider the correlation of the high-dimensional features, and the LSTM network is used to consider the temporal correlation of traffic flow data. Moreover, the multi-head attention mechanism is introduced to use the most relevant portion of the traffic data to make predictions so that the prediction performance can be improved. A traffic flow dataset collected from the Caltrans Performance Measurement System (PeMS) database is used to demonstrate the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Objective: Accurate segmentation of liver tumors, which could help physicians make appropriate treatment decisions and assess the effectiveness of surgical treatment, is crucial for the clinical diagnosis of liver cancer. In this study, we propose a 4-dimensional (4D) deep learning model based on 3D convolution and convolutional long short-term memory (C-LSTM) for hepatocellular carcinoma (HCC) lesion segmentation. Methods: The proposed deep learning model utilizes 4D information on dynamic contrast enhanced (DCE) magnetic resonance imaging (MRI) images to assist liver tumor segmentation. Specifically, a shallow U-net based 3D CNN module was designed to extract 3D spatial domain features from each DCE phase, followed by a 4-layer C-LSTM network module for time domain information exploitation. The combined information of multi-phase DCE images and the manner by which tissue imaging features change on multi-contrast images allow the network to more effectively learn the characteristics of HCC, resulting in better segmentation performance. Results: The proposed model achieved a Dice score of 0.825± 0.077, a Hausdorff distance of 12.84± 8.14 mm, and a volume similarity of 0.891± 0.080 for liver tumor segmentation, which outperformed the 3D U-net model, RA-UNet model and other models in the ablation study in both internal and external test sets. Moreover, the performance of the proposed model is comparable to the nnU-Net model, which showed state-of-the-art performance in many segmentation tasks, with significantly reduced prediction time. Conclusion: The proposed 3D convolution and C-LSTM based model can achieve accurate segmentation of HCC lesions. [ABSTRACT FROM AUTHOR]

Diabetic Retinopathy is one of the prominent reasons for permanent blindness in working age, long term diabetic patients. With the prevalence in raise of diabetics, majority of the people are endangered to permanent vision loss. The advancements in medical imaging techniques enabled the research community to focus on developing automated and computerized systems for diagnosing retinopathy in early stages. But, it is a very complex challenge due to the presence of high intra-class variations and imbalanced data distribution for higher grades of severity. In recent years, various deep learning based models have been designed for automating the process of retinopathy severity classification. In this research work, we present a fascinating deep learning model with multiple attention stages called Hinge Attention Network (HA-Net). Proposed model consists of a pre-trained VGG16 base to extract initial spatial representation from retinal scan images, spatial attention autoencoder to learn lesion specific latent representations in spatial dimensions and a channel attention based hinge neural network to grab category based discriminative features in channel dimension and classify the severity grade of retinopathy. In addition to spatial and channel attention mechanism, we use Convolutional LSTM layer to prioritize highly important spatial maps before passing to hinge neural network. All these components of HA-Net, enabled it to make generalised and accurate predictions on unseen data. The effectiveness and acceptability of proposed model is proved by validating it using two benchmark datasets, Kaggle APTOS 2019 and ISBI IDRiD. Extensive experimental studies on these datasets reveal that, proposed HA-Net outstrip several existing models by achieving an accuracy of 85.54% on Kaggle APTOS, and an accuracy of 66.41% on IDRiD datasets. [ABSTRACT FROM AUTHOR]

Abstract Trajectory prediction is a fundamental and challenging task for numerous applications, such as autonomous driving and intelligent robots. Current works typically treat pedestrian trajectories as a series of 2D point coordinates. However, in real scenarios, the trajectory often exhibits randomness, and has its own probability distribution. Inspired by this observation and other movement characteristics of pedestrians, we propose a simple and intuitive movement description called a trajectory distribution, which maps the coordinates of the pedestrian trajectory to a 2D Gaussian distribution in space. Based on this novel description, we develop a new trajectory prediction method, which we call the social probability method. The method combines trajectory distributions and powerful convolutional recurrent neural networks. Both the input and output of our method are trajectory distributions, which provide the recurrent neural network with sufficient spatial and random information about moving pedestrians. Furthermore, the social probability method extracts spatio-temporal features directly from the new movement description to generate robust and accurate predictions. Experiments on public benchmark datasets show the effectiveness of the proposed method.

It is of great significance to accurately grasp the demand for air travel to promote the revival of long-distance travel and alleviate public anxiety. The main purpose of this study is to build a high-precision air travel demand forecasting framework by introducing effective Internet data. In the age of big data, passengers before traveling often look for reference groups in search engines and make travel decisions under their informational influence. The big data generated based on these behaviors can reflect the overall passenger psychology and travel demand. Therefore, based on big data mining technology, this study designed a strict dual data preprocessing method and an ensemble forecasting framework, introduced search engine data into the air travel demand forecasting process, and conducted empirical research based on the dataset composed of air travel volume of Shanghai Pudong International Airport. The results show that effective search engine data is helpful to air travel demand forecasting. This research provides a theoretical basis for the application of big data mining technology and data spatial information in air travel demand forecasting and tourism management, and provides a new idea for alleviating public anxiety.

Demand prediction for on-demand food delivery (ODFD) is of great importance to the operation and transportation resource utilization of ODFD platforms. This paper addresses short-term ODFD demand prediction using an end-to-end deep learning architecture. The problem is formulated as a spatial–temporal prediction. The proposed model is composed of convolutional long short-term memory (ConvLSTM), and convolutional neural network (CNN) units with encoder–decoder structure. Specifically, long short-term memory (LSTM) networks are a type of recurrent neural network capable of learning order dependence in sequence prediction problems. The convolution unit is responsible for capturing spatial attributes, while the LSTM part is adopted to learn temporal attributes. Additionally, an attentional model is designed and integrated to enhance the prediction performance by addressing the spatial variation in demand. The proposed approach is compared to several baseline models using a historical ODFD dataset from Shenzhen, China. Results indicate that the proposed model obtains the highest prediction accuracy by capturing both spatial and temporal correlations with attention information focusing on different parts of the input series.

The accurate detection and recognition of human actions play a pivotal role in aerial surveillance, enabling the identification of potential threats and suspicious behavior. Several approaches have been presented to address this problem, but the limitation still remains in devising an accurate and robust solution. To this end, this paper presents an effective action recognition framework for aerial surveillance, employing the YOLOv8-Pose keypoints extraction algorithm and a customized sequential ConvLSTM (Convolutional Long Short-Term Memory) model for classifying the action. We performed a detailed experimental evaluation and comparison on the publicly available Drone Action dataset. The evaluation and comparison of the proposed framework with several existing approaches on the publicly available Drone Action dataset demonstrate its effectiveness, achieving a very encouraging performance. The overall accuracy of the framework on three provided dataset splits is 74%, 80%, and 70%, with a mean accuracy of 74.67%. Indeed, the proposed system effectively captures the spatial and temporal dynamics of human actions, providing a robust solution for aerial action recognition.

Land surface temperature (LST) prediction is of great importance for climate change, ecology, environmental and industrial studies. These studies require accurate LST map predictions considering both spatial and temporal dynamics. In this study, multilayer perceptron (MLP), long short-term memory (LSTM) and an integrated machine learning model, namely Convolutional LSTM (ConvLSTM), were utilized for one step ahead LST prediction. Data were gathered from 1-day (MYD11A1) and 8-day composite (MYD11A2) Moderate Resolution Imaging Spectroradiometer (MODIS) sensors, which have 1-km × 1-km spatial resolution. Considering the inability of MODIS sensors to provide LST data under cloudy conditions, Inverse DISTANCE WEIGHTING (IDW), natural neighbor (NN), and cubic spline (C) methods were used to overcome the missing pixel problem. The proposed methods were tested over the Northern part of Adana province, Turkey, and the performances of the models were quantitatively evaluated through performance measures, namely, Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). The selected datasets range from 01 January 2017 to 01 November 2020 and from 01 January 2015 to 01 November 2020 for daily LST and 8-day composite LST, respectively. While 60% of the datasets were used as training set, the remaining 40% were used as validation (20%) and test (20%) sets. RMSE maps were generated to evaluate the pixelwise performance of the proposed method. On the other hand, the best average RMSE and MAE for the daily test set were obtained from the combination of ConvLSTM and NN (NN-ConvLSTM) as 3.62 °C and 2.85 °C, respectively, while they were acquired 3.57 °C and 2.69 oC from the combination of MLP and NN (NN-MLP) for the 8-day composite LST test set. The results revealed that the proposed hybrid models could be used for one step ahead spatiotemporal prediction of LST data. [ABSTRACT FROM AUTHOR]