Sliding Window Size Research Articles

Ubiquitous Transportation Mode Detection Using Single Cell Tower Information

Detecting the transportation mode of a user has an overwhelming significance for a vast domain of applications. It has been addressed by several systems using ubiquitous mobile phones. These systems leverage multiple cell towers' information, GPS, and inertial sensors. However, these phone sensors are limited to a small subset of phones (e.g. high-end phones or phones that support neighboring cell towers information), have high energy consumption, cannot work in certain areas (e.g. inside tunnels for GPS), and/or work only from the user side. Furthermore, there are variations in the receiving capabilities of different cell phone devices according to their brand, quality of receiver and cost which raises the problem of device heterogeneity. In this paper, a transportation mode detection system, MonoSense, is presented that leverages the phone serving cell information only. The basic idea is that the phone speed can be correlated with features extracted directly from both the serving cell tower ID and the received signal strength (RSS) from it. To achieve high detection accuracy with this limited information MonoSense leverages diversity along multiple feature dimensions to synthesize novel features. Specifically, MonoSense exploits both the time and frequency information available from the serving cell tower over different sliding window sizes. Besides, it is shown that both the logarithmic and linear RSS scales provide useful information about the movement of a phone, further enriching the feature space and leading to higher accuracy. Moreover, to overcome the heterogeneity problem, a set of device-independent features is derived and added to the feature set. MonoSense has been evaluated using 500 hours of cellular traces covering 1500 km for more than sixteen months and collected by four users using different android phones. MonoSense achieved an average precision, recall and F-measure of 87.3%, 85.9%, and 86.4%, respectively, in discriminating between stationary, walking and driving modes using only the serving cell tower information, highlighting its ability to enable a wide set of intelligent transportation applications.

Automated pavement distress detection using region based convolutional neural networks

ABSTRACT Automatic pavement crack detection is essential for evaluating maintenance requirements and ensuring driving safety. Crack detection plays a primary role in realising the automatic evaluation of pavement condition. Most existing researches on pavement crack detection rely on laborious work, which is a time- and cost-intensive process. Although there has been considerable research on pavement crack detection, it remains a challenging task owing to diverse complex pavement conditions. Recently, deep learning-based algorithms have achieved significant success in computer vision tasks. However, the techniques still have limitations for automatic pavement distress detection. To overcome the current limitations, this study proposes a method for detecting signs of pavement distress based on faster region based convolutional neural network (Faster R-CNN). The study focuses on the detection of longitudinal cracks, transverse cracks, alligator cracks, and partial patching in pavement images. A framework for applying the Faster R-CNN technique to a full-size pavement image is also proposed, which allows the sliding window size to be reduced, thus enabling the detection of larger images. The performance of the proposed method was validated against a dataset containing actual pavement images. The experimental test results show that the proposed method could successfully detect cracks and partial patching with accuracy.

Parameters optimization of the Structural Similarity Index

We exploit evolutionary computation to optimize the handcrafted Structural Similarity method (SSIM) through a datadriven approach. We estimate the best combination of luminance, contrast and structure components, as well as the sliding window size used for processing, with the objective of optimizing the similarity correlation with human-expressed mean opinion score on a standard dataset. We experimentally observe that better results can be obtained by penalizing the overall similarity only for very low levels of luminance similarity. Finally, we report a comparison of SSIM with the optimized parameters against other metrics for full reference quality assessment, showing superior performance on a different dataset.

Convolutional Sparse Coded Dynamic Brain Functional Connectivity

Functional brain network has been widely studied in many previous work for brain disorder diagnosis and brain network analysis. However, most previous work focus on static dynamic brain network research. Lots of recent work reveals that the brain shows dynamic activity even in resting state. Such dynamic brain functional connectivity reveals discriminative patterns for identifying many brain disorders. Current sliding window based dynamic brain connectivity framework are not easy to be applied to real clinical applications due to many issues: First, how to set up the optimal sliding window size and how to determine the threshold for the brain connectivity patterns. Secondly, how to represent the high dimensional dynamic brain connectivity pattern in a low dimensional representations for diagnosis purpose. Last, how to deal with the different length dynamic brain network patterns especially when the raw data are of different length. In order to address all those above issues, we proposed a new framework, which employs multiple scale sliding windows and automatically learns a sparse and low ran dynamic brain functional connectivity patterns from raw fMRI data. Furthermore, we are able to measure different length dynamic brain functional connectivity patterns in an equal space by learning a sparse coded convolutional filters. We have evaluated our method with state of the art dynamic brain network methods and the results demonstrated the strong potential of our methods for brain disorder diagnosis in real clinical applications.

A Real-Time Capable Linear Time Classifier Scheme for Anticipated Hand Movements Recognition from Amputee Subjects Using Surface EMG Signals

Anticipated hand movements of amputee subjects are considered difficult to classify using only Electromyogram (EMG) signals and machine learning techniques. For a long time, classifying such s-EMG signals have been considered as a non-linear problem, and the problem of signal sparsity has not been given detailed attention in a large set of action classes. For addressing these problems, this paper is proposing a linear-time classifier termed as Random Fourier Mapped Collaborative Representation with distance weighted Tikhonov regularization matrix (RFMCRT). RFMCRT attempts to tackle the non-linear problem via Random Fourier Features and sparsity issue with collaborative representation. The projection error of Random Fourier Features is reduced by projecting to the same dimension as the original feature space and later finding the collaborative representation, with an optional non-negative constraint (RFMNNCRT). The proposed two classifiers were tested with time-domain features computed from the EMG signals obtained from NINAPRO databases using a non-overlapping sliding window size of 256 ms. Due to the random nature of our proposed classifiers, this paper has computed the average and worst-case performance for 50 trials and compared them with other reported classifiers. The results show that RFMNNCRT (average case) outperformed state-of-the-art classifiers with the accuracy of 93.44% for intact subjects and 55.67% for amputee subjects. In the worst-case situation, RFMCRT achieves considerable performance for the same, with the reported accuracy of 91.55% and 50.27% respectively. Our proposed classifier guarantees acceptable levels of accuracy for large classes of hand movements and also maintains good computational efficiency in comparison to LDA and SVM.

Mining frequent itemsets from streaming transaction data using genetic algorithms

This paper presents a study of mining frequent itemsets from streaming data in the presence of concept drift. Streaming data, being volatile in nature, is particularly challenging to mine. An approach using genetic algorithms is presented, and various relationships between concept drift, sliding window size, and genetic algorithm constraints are explored. Concept drift is identified by changes in frequent itemsets. The novelty of this work lies in determining concept drift using frequent itemsets for mining streaming data, using the genetic algorithm framework. Formulas have been presented for calculating minimum support counts in streaming data using sliding windows. Testing highlighted that the ratio of the window size to transactions per drift was a key to good performance. Getting good results when the sliding window size was too small was a challenge since normal fluctuations in the data could appear to be a concept drift. Window size must be managed in conjunction with support and confidence values in order to achieve reasonable results. This method of detecting concept drift performed well when larger window sizes were used.

Monte Carlo Optimization for Sliding Window Size in Dixon Quality Control of Environmental Monitoring Time Series Data

Outliers are often present in large datasets of water quality monitoring time series data. A method of combining the sliding window technique with Dixon detection criterion for the automatic detection of outliers in time series data is limited by the empirical determination of sliding window sizes. The scientific determination of the optimal sliding window size is very meaningful research work. This paper presents a new Monte Carlo Search Method (MCSM) based on random sampling to optimize the size of the sliding window, which fully takes advantage of computers and statistics. The MCSM was applied in a case study to automatic monitoring data of water quality factors in order to test its validity and usefulness. The results of comparing the accuracy and efficiency of the MCSM show that the new method in this paper is scientific and effective. The experimental results show that, at different sample sizes, the average accuracy is between 58.70% and 75.75%, and the average computation time increase is between 17.09% and 45.53%. In the era of big data in environmental monitoring, the proposed new methods can meet the required accuracy of outlier detection and improve the efficiency of calculation.

Exploiting the local temporal information for video captioning

Typical video captioning methods are developed based on the encoder-decoder architecture. To better exploit the local temporal information, e.g., details about objects and their corresponding actions, we propose a reinforcement learning based method to predict the adaptive sliding window size sequentially for better event exploration. More specifically, we introduce the single Monte-Carlo sample to approximate the gradient of reward-based loss function. And the self-critical strategy is employed to estimate baseline reward to diminish the variance of gradients. Moreover, temporal attention is utilized to selectively focus on a subset of temporal frame representations while generating each word. In addition, to better initialize the decoder’s state, we utilize the motion features extracted by 3D CNNs with mean pooling to endow the decoder with the prior knowledge of the entire video. To evaluate the proposed method, experiments are performed on three public benchmark datasets: Microsoft Video Description Corpus (MSVD), MSR Video to Text challenge (MSR-VTT) and Charades. The experimental results demonstrate the effectiveness of our method by comparing with state-of-the-art methods.

Design of Smart Home Implementation Within IoT Natural Language Interface

To process continuous sensor data in Internet of Things (IoT) environments, this study optimizes queries using multiple MJoin operators. To achieve efficient storage management, it classifies and reduces data using a support vector machine (SVM) classification algorithm. A global shared query execution technique was used to optimize multiple MJoin queries. By comparing each kernel function of the SVM classification algorithm, the system's performance was evaluated through experiments according to the selected optimal kernel function and changes in sliding window size. Furthermore, to implement a smart home system that can actively respond to users, classified and reduced sensor data were utilized to enable the intelligent control of devices inside the home. The sensor data (e.g., temperature, humidity, gas) used to recognize the current conditions of an IoT-based smart home system and corresponding date data were classified into decision trees, and the system was designed using five sensors to intelligently control priorities such as ventilation, temperature, and fire and intrusion detection. The experiments demonstrated that the multiple MJoin technique yields high improvements in performance with relatively few searches. In this study, the sigmoid was selected as the optimal kernel function for the SVM classification algorithm. According to the SVM classification algorithm results, based on changes in the sliding window size, the average error rate was 2.42%, the reduction result was 17.58%, and the classification accuracy was 85.94%. According to the comparison of the classification performance of SVM and other algorithms, the SVM classification algorithm exhibited a minimum 9% better classification performance. Thus, compared to existing home systems, this algorithm is expected to increase system efficiency and convenience by enabling the configuration of a more intelligent environment according to the user's characteristics or requirements.

Oil Palm Tree Detection from High Resolution Drone Image Using Convolutional Neural Network

We propose a method to combine several image processing methods with Convolutional Neural Network (CNN) to perform palm tree detection and counting. This paper focuses on drone imaging, which has a high image resolution and is widely deployed in the plantation industry. Analyzing drone images is challenging due to variable drone flying altitudes, resulting in inconsistent tree sizes in images captured. Counting by template matching or fixed sliding window size method often produces an inaccurate count. Instead, our method employs frequency domain analysis to estimate tree size before CNN. The method is evaluated using two images, ranging from a few thousand trees to a few hundred thousand trees per image. We have summarized the accuracy of the proposed method by comparing the results with manually labelled ground truth.

Oil Palm Tree Detection from High Resolution Drone Image Using Convolutional Neural Network

We propose a method to combine several image processing methods with Convolutional Neural Network (CNN) to perform palm tree detection and counting. This paper focuses on drone imaging, which has a high image resolution and is widely deployed in the plantation industry. Analyzing drone images is challenging due to variable drone flying altitudes, resulting in inconsistent tree sizes in images captured. Counting by template matching or fixed sliding window size method often produces an inaccurate count. Instead, our method employs frequency domain analysis to estimate tree size before CNN. The method is evaluated using two images, ranging from a few thousand trees to a few hundred thousand trees per image. We have summarized the accuracy of the proposed method by comparing the results with manually labelled ground truth.

Retrospective deformation of the Baige landslide using optical remote sensing images

Detecting slope deformations before major landslides is crucially important for mitigating landslide disasters. In this work, we explored the use of Sentinel-2 images and an image correlation method, Co-Registration of Optically Sensed Images and Correlation (COSI-Corr), to derive slope deformations before the 2018 Baige landslide along the Jinsha River. We found that larger sliding windows outperform smaller ones in extracting the moving slope. Different solar zenith angles in image pairs have an influence on the derived displacements. Although the spatial resolution of Sentinel-2 images is 10 m, uncertainties in estimating slope deformation are found within 2 m. Using the COSI-Corr technique, we monitored sub-pixel slope deformation in Sentinel-2 images from November 2015 to August 2018. We found that the slope moved faster in summer than in winter months, which was probably driven by the monsoon climate in this region. From 2015 to 2018, the slope movement became faster, and an acceleration stage was observed before the slope failed completely. To derive reliable slope displacements, this work suggests testing different sliding window size combinations in the use of image correlation methods, and optical images acquired with similar zenith angles are preferred. This work also demonstrated the capability of using optical remote sensing images to monitor large slope deformations before major landslides. As more optical images with high spatial resolution become available, monitoring minor slope deformations directly would be practicable for large remote mountain regions.

Adaptive sliding windows for improved estimation of data center resource utilization

Accurate prediction of data center resource utilization is required for capacity planning, job scheduling, energy saving, workload placement, and load balancing to utilize the resources efficiently. However, accurately predicting those resources is challenging due to dynamic workloads, heterogeneous infrastructures, and multi-tenant co-hosted applications. Existing prediction methods use fixed size observation windows which cannot produce accurate results because of not being adaptively adjusted to capture local trends in the most recent data. Therefore, those methods train on large fixed sliding windows using an irrelevant large number of observations yielding to inaccurate estimations or fall for inaccuracy due to degradation of estimations with short windows on quick changing trends. In this paper we propose a deep learning-based adaptive window size selection method, dynamically limiting the sliding window size to capture the trend for the latest resource utilization, then build an estimation model for each trend period. We evaluate the proposed method against multiple baseline and state-of-the-art methods, using real data-center workload data sets. The experimental evaluation shows that the proposed solution outperforms those state-of-the-art approaches and yields 16 to 54% improved prediction accuracy compared to the baseline methods.

A bidirectional weighted boundary distance algorithm for time series similarity computation based on optimized sliding window size

<p style='text-indent:20px;'>The existing method of determining the size of the time series sliding window by empirical value exists some problems which should be solved urgently, such as when considering a large amount of information and high density of the original measurement data collected from industry equipment, the important information of the data cannot be maximally retained, and the calculation complexity is high. Therefore, by studying the effect of sliding window on time series similarity technology in practical application, an algorithm to determine the initial size of the sliding window is proposed. The upper and lower boundary curves with a higher fitting degree are constructed, and the trend weighting is introduced into the <inline-formula><tex-math id="M1">\begin{document}$ LB\_Hust $\end{document}</tex-math></inline-formula> distance calculation method to reduce the difficulty of mathematical modeling and improve the efficiency of data similarity computation.

Highly accurate and explainable detection of specimen mix-up using a machine learning model.

Background Delta check is widely used for detecting specimen mix-ups. Owing to the inadequate specificity and sparseness of the absolute incidence of mix-ups, the positive predictive value (PPV) of delta check is considerably low as it is labor consuming to identify true mix-up errors among a large number of false alerts. To overcome this problem, we developed a new accurate detection model through machine learning. Methods Inspired by delta check, we decided to conduct comparisons with the past examinations and broaden the time range. Fifteen common items were selected from complete blood cell counts and biochemical tests. We considered examinations in which ≥11 among the 15 items were measured simultaneously in our hospital; we created individual partial time-series data of the consecutive examinations with a sliding window size of 4. The last examinations of the partial time-series data were shuffled to generate artificial mix-up cases. After splitting the dataset into development and validation sets, we allowed a gradient-boosting-decision-tree (GBDT) model to learn using the development set to detect whether the last examination results of the partial time-series data were artificial mixed-up results. The model's performance was evaluated on the validation set. Results The area under the receiver operating characteristic curve (ROC AUC) of our model was 0.9983 (bootstrap confidence interval [bsCI]: 0.9983-0.9985). Conclusions The GBDT model was more effective in detecting specimen mix-up. The improved accuracy will enable more facilities to perform more efficient and centralized mix-up detection, leading to improved patient safety.

Remaining useful life prediction of lithium-ion batteries based on false nearest neighbors and a hybrid neural network

Accurate estimation of the remaining useful life of lithium-ion batteries is critically important for electronic devices. In the existing literature, the widely applied model-based approaches for remaining useful battery life estimation are limited by the complexity of the electrochemical modeling required. In addition, data-driven approaches for remaining useful battery life estimation commonly define unreliable sliding window sizes empirically and the prediction accuracy of these approaches needs to be improved. To address the above issues, use of a hybrid neural network with the false nearest neighbors method is proposed in this paper. First, the false nearest neighbors method is used to calculate the sliding window size required for prediction. Second, a hybrid neural network that combines the advantages of a convolutional neural network with those of long short-term memory is designed for model training and prediction. Remaining useful life prediction experiments for batteries with various rated capacities are performed to verify the effectiveness of the proposed approach, and the results demonstrate that the proposed approach offers wide generality and reduced errors when compared with the other state-of-the-art methods.

Neural Offset Time Evaluation in Surface Respiratory Signals during Controlled Respiration.

The electrical activity of the diaphragm measured by surface electromyography (sEMGdi) provides indirect information on neural respiratory drive. Moreover, it allows evaluating the ventilatory pattern from the onset and offset (ntoff) estimation of the neural inspiratory time. sEMGdi amplitude variation was quantified using the fixed sample entropy (fSampEn), a less sensitive method to the interference from cardiac activity. The detection of the ntoff is controversial, since it is located in an intermediate point between the maximum value and the cessation of sEMGdi inspiratory activity, evaluated by the fSampEn. In this work ntoff detection has been analyzed using thresholds between 40% and 100 % of the fSampEn peak. Furthermore, fSampEn was evaluated analyzing the r parameter from 0.05 to 0.6, using a m equal to 1 and a sliding window size equal to 250 ms. The ntoff has been compared to the offset time (toff) obtained from the airflow during a controlled respiratory protocol varying the fractional inspiratory time from 0.54 to 0.18 whilst the respiratory rate was constant at 16 bpm. Results show that the optimal threshold values were between 66.0 % to 77.0 % of the fSampEn peak value. r values between 0.25 to 0.50 were found suitable to be used with the fSampEn.

Multiscale running correlation analysis of water quality datasets of Noyyal River, India, using the Hilbert–Huang Transform

This study performs the multiscale decomposition of six water quality parameters from Elunuthimangalam station in Noyyal River, a water quality hotspot in Southern India, using the complete ensemble empirical mode decomposition with adaptive noise method. Then, the intrinsic mode functions (IMFs) obtained are subjected to normalized Hilbert transform-direct quadrature-coupled framework for their time–frequency characterization. The time–frequency–amplitude spectra revealed that the dominant frequency is dynamic in characteristics and the marginal spectra successfully captured the significant high anthropogenic interventions in the form of pollutant disposals in the study area. Then, an in-depth examination of the association of different water quality parameters such as pH, temperature, dissolved oxygen (DO), biochemical oxygen demand (BOD) and total hardness (TH) with electrical conductivity (EC) is done through a running correlation method, namely time-dependent intrinsic correlation (TDIC) in which the sliding window size is adaptively fixed based on instantaneous frequencies estimated by HT. The TDIC analysis revealed that with the exception of TH, the association of water quality parameters with EC in different time scales is not alike in both nature and strength. Also the well-debated DO–temperature and DO–BOD relationships displayed diverse correlation properties in different time scales and over the time domain.

Applying Catastrophe Theory for Network Anomaly Detection in Cloud Computing Traffic

In spite of the tangible advantages of cloud computing, it is still vulnerable to potential attacks and threats. In light of this, security has turned into one of the main concerns in the adoption of cloud computing. Therefore, an anomaly detection method plays an important role in providing a high protection level for network security. One of the challenges in anomaly detection, which has not been seriously considered in the literature, is applying the dynamic nature of cloud traffic in its prediction while maintaining an acceptable level of accuracy besides reducing the computational cost. On the other hand, to overcome the issue of additional training time, introducing a high-speed algorithm is essential. In this paper, a network traffic anomaly detection model grounded in Catastrophe Theory is proposed. This theory is effective in depicting sudden change processes of the network due to the dynamic nature of the cloud. Exponential Moving Average (EMA) is applied for the state variable in sliding window to better show the dynamicity of cloud network traffic. Entropy is used as one of the control variables in catastrophe theory to analyze the distribution of traffic features. Our work is compared with Wei Xiong et al.’s Catastrophe Theory and achieved a maximum improvement in the percentage of Detection Rate in week 4 Wednesday (7.83%) and a 0.31% reduction in False Positive Rate in week 5 Monday. Additional accuracy parameters are checked and the impact of sliding window size in sensitivity and specificity is considered.

Public Transport Driver Identification System Using Histogram of Acceleration Data

This paper introduces a driver identification system architecture for public transport which utilizes only acceleration sensor data. The system architecture consists of three main modules which are the data collection, data preprocessing, and driver identification module. Data were collected from real operation of campus shuttle buses. In the data preprocessing module, a filtering module is proposed to remove the inactive period of the public transport data. To extract the unique behavior of the driver, a histogram of acceleration sensor data is proposed as a main feature of driver identification. The performance of our system is evaluated in many important aspects, considering axis of acceleration, sliding window size, number of drivers, classifier algorithms, and driving period. Additionally, the case study of impostor detection is implemented by modifying the driver identification module to identify a car thief or carjacking. Our driver identification system can achieve up to 99% accuracy and the impostor detection system can achieve the F1 score of 0.87. As a result, our system architecture can be used as a guideline for implementing the real driver identification system and further driver identification researches.