Sensors

Research

Jump to: Other

22 pages, 828 KiB

Open AccessArticle

Measuring the Uncertainty of Predictions in Deep Neural Networks with Variational Inference

by Jan Steinbrener, Konstantin Posch and Jürgen Pilz

Sensors 2020, 20(21), 6011; https://doi.org/10.3390/s20216011 - 23 Oct 2020

Cited by 13 | Viewed by 4301

Abstract

We present a novel approach for training deep neural networks in a Bayesian way. Compared to other Bayesian deep learning formulations, our approach allows for quantifying the uncertainty in model parameters while only adding very few additional parameters to be optimized. The proposed [...] Read more.

We present a novel approach for training deep neural networks in a Bayesian way. Compared to other Bayesian deep learning formulations, our approach allows for quantifying the uncertainty in model parameters while only adding very few additional parameters to be optimized. The proposed approach uses variational inference to approximate the intractable a posteriori distribution on basis of a normal prior. By representing the a posteriori uncertainty of the network parameters per network layer and depending on the estimated parameter expectation values, only very few additional parameters need to be optimized compared to a non-Bayesian network. We compare our approach to classical deep learning, Bernoulli dropout and Bayes by Backprop using the MNIST dataset. Compared to classical deep learning, the test error is reduced by 15%. We also show that the uncertainty information obtained can be used to calculate credible intervals for the network prediction and to optimize network architecture for the dataset at hand. To illustrate that our approach also scales to large networks and input vector sizes, we apply it to the GoogLeNet architecture on a custom dataset, achieving an average accuracy of 0.92. Using 95% credible intervals, all but one wrong classification result can be detected. Full article

(This article belongs to the Special Issue Data Analysis for Smart Sensor Systems)

► Show Figures

Figure 1

19 pages, 5790 KiB

Open AccessArticle

Knowledge Discovery Using Topological Analysis for Building Sensor Data

by Manik Gupta and Nigel Phillips

Sensors 2020, 20(17), 4914; https://doi.org/10.3390/s20174914 - 31 Aug 2020

Cited by 3 | Viewed by 2412

Abstract

Distributed sensor networks are at the heart of smart buildings, providing greater detail and valuable insights into their energy consumption patterns. The problem is particularly complex for older buildings retrofitted with Building Energy Management Systems (BEMS) where extracting useful knowledge from large sensor [...] Read more.

Distributed sensor networks are at the heart of smart buildings, providing greater detail and valuable insights into their energy consumption patterns. The problem is particularly complex for older buildings retrofitted with Building Energy Management Systems (BEMS) where extracting useful knowledge from large sensor data streams without full understanding of the underlying system variables is challenging. This paper presents an application of Q-Analysis, a computationally simple topological approach for summarizing large sensor data sets and revealing useful relationships between different variables. Q-Analysis can be used to extract novel structural features called Q-vectors. The Q-vector magnitude visualizations are shown to be very effective in providing insights on macro behaviors, i.e., building floor behaviors in the present case, which are not evident from the use of unsupervised learning algorithms applied on individual terminal units. It has been shown that the building floors exhibited distinct behaviors that are dependent on the set-point distribution, but independent of the time and season of the year. Full article

(This article belongs to the Special Issue Data Analysis for Smart Sensor Systems)

► Show Figures

Figure 1

20 pages, 5435 KiB

Open AccessArticle

Uncertainty Principle and Power Quality Sensing and Analysis in Smart Substation

by Wanli Cai, Li Wu, Yibo Cui and Shunfan He

Sensors 2020, 20(15), 4281; https://doi.org/10.3390/s20154281 - 31 Jul 2020

Cited by 3 | Viewed by 2900

Abstract

Different kinds of power quality can be sensed in a smart substation. Power quality sensing and analysis are basic functions of a smart substation for situation awareness. The uncertainty principle, which states that the time uncertainty and frequency uncertainty cannot be minimized simultaneously, [...] Read more.

Different kinds of power quality can be sensed in a smart substation. Power quality sensing and analysis are basic functions of a smart substation for situation awareness. The uncertainty principle, which states that the time uncertainty and frequency uncertainty cannot be minimized simultaneously, is a bottleneck problem that undermines the faithfulness of sensing and confines the accuracy of analysis. This paper studies the influence of the uncertainty principle on the power quality monitoring issue in detail and solves the problem by ideal atomic decomposition (IAD). The new method employs a pair of time and frequency bases where the power quality waveform is sensed. Then, both time uncertainty and frequency uncertainty can be minimized simultaneously. The sensing process is realized by orthogonal matching pursuit (OMP). By simulated and field power quality tests with comparisons of developed methods, the new method can give faithful sensing and accurate analysis for various power qualities, and is validated as an effective power quality monitoring method in smart substations. Full article

(This article belongs to the Special Issue Data Analysis for Smart Sensor Systems)

► Show Figures

Figure 1

22 pages, 6083 KiB

Open AccessArticle

Deep Learning for Feature Extraction in Remote Sensing: A Case-Study of Aerial Scene Classification

by Biserka Petrovska, Eftim Zdravevski, Petre Lameski, Roberto Corizzo, Ivan Štajduhar and Jonatan Lerga

Sensors 2020, 20(14), 3906; https://doi.org/10.3390/s20143906 - 14 Jul 2020

Cited by 78 | Viewed by 8417

Abstract

Scene classification relying on images is essential in many systems and applications related to remote sensing. The scientific interest in scene classification from remotely collected images is increasing, and many datasets and algorithms are being developed. The introduction of convolutional neural networks (CNN) [...] Read more.

Scene classification relying on images is essential in many systems and applications related to remote sensing. The scientific interest in scene classification from remotely collected images is increasing, and many datasets and algorithms are being developed. The introduction of convolutional neural networks (CNN) and other deep learning techniques contributed to vast improvements in the accuracy of image scene classification in such systems. To classify the scene from areal images, we used a two-stream deep architecture. We performed the first part of the classification, the feature extraction, using pre-trained CNN that extracts deep features of aerial images from different network layers: the average pooling layer or some of the previous convolutional layers. Next, we applied feature concatenation on extracted features from various neural networks, after dimensionality reduction was performed on enormous feature vectors. We experimented extensively with different CNN architectures, to get optimal results. Finally, we used the Support Vector Machine (SVM) for the classification of the concatenated features. The competitiveness of the examined technique was evaluated on two real-world datasets: UC Merced and WHU-RS. The obtained classification accuracies demonstrate that the considered method has competitive results compared to other cutting-edge techniques. Full article

(This article belongs to the Special Issue Data Analysis for Smart Sensor Systems)

► Show Figures

Figure 1

23 pages, 4593 KiB

Open AccessArticle

An Intelligent Multi-Sensor Variable Spray System with Chaotic Optimization and Adaptive Fuzzy Control

by Lepeng Song, Jinpen Huang, Xianwen Liang, Simon X. Yang, Wenjin Hu and Dedong Tang

Sensors 2020, 20(10), 2954; https://doi.org/10.3390/s20102954 - 22 May 2020

Cited by 17 | Viewed by 3444

Abstract

During the variable spray process, the micro-flow control is often held back by such problems as low initial sensitivity, large inertia, large hysteresis, nonlinearity as well as the inevitable difficulties in controlling the size of the variable spray droplets. In this paper, a [...] Read more.

During the variable spray process, the micro-flow control is often held back by such problems as low initial sensitivity, large inertia, large hysteresis, nonlinearity as well as the inevitable difficulties in controlling the size of the variable spray droplets. In this paper, a novel intelligent double closed-loop control with chaotic optimization and adaptive fuzzy logic is developed for a multi-sensor based variable spray system, where a Bang-Bang relay controller is used to speed up the system operation, and adaptive fuzzy nonlinear PID is employed to improve the accuracy and stability of the system. With the chaotic optimization of controller parameters, the system is globally optimized in the whole solution space. By applying the proposed double closed-loop control, the variable pressure control system includes the pressure system as the inner closed-loop and the spray volume system as the outer closed-loop. Thus, the maximum amount of spray droplets deposited on the plant surface may be achieved with the minimum medicine usage for plants. Multiple sensors (for example: three pressure sensors and two flow rate sensors) are employed to measure the system states. Simulation results show that the chaotic optimized controller has a rise time of 0.9 s, along with an adjustment time of 1.5 s and a maximum overshoot of 2.67% (in comparison using PID, the rise time is 2.2 s, the adjustment time is 5 s, and the maximum overshoot is 6.0%). The optimized controller parameters are programmed into the hardware to control the established variable spray system. The experimental results show that the optimal spray pressure of the spray system is approximately 0.3 MPa, and the flow rate is approximately 0.08 m³/h. The effective droplet rate is 89.4%, in comparison to 81.3% using the conventional PID control. The proposed chaotically optimized composite controller significantly improved the dynamic performance of the control system, and satisfactory control results are achieved. Full article

(This article belongs to the Special Issue Data Analysis for Smart Sensor Systems)

► Show Figures

Figure 1

17 pages, 3280 KiB

Open AccessArticle

Dynamically Reconfigurable Data Readout of Pixel Detectors for Automatic Synchronization with Data Acquisition Systems

by Farah Fahim, Simone Bianconi, Jacob Rabinowitz, Siddhartha Joshi and Hooman Mohseni

Sensors 2020, 20(9), 2560; https://doi.org/10.3390/s20092560 - 30 Apr 2020

Cited by 4 | Viewed by 3402

Abstract

Reconfigurable detectors with dynamically selectable sensing and readout modes are highly desirable for implementing edge computing as well as enabling advanced imaging techniques such as foveation. The concept of a camera system capable of simultaneous passive imaging and dynamic ranging in different regions [...] Read more.

Reconfigurable detectors with dynamically selectable sensing and readout modes are highly desirable for implementing edge computing as well as enabling advanced imaging techniques such as foveation. The concept of a camera system capable of simultaneous passive imaging and dynamic ranging in different regions of the detector is presented. Such an adaptive-autonomous detector with both spatial and temporal control requires programmable window of exposure (time frames), ability to switch between readout modes such as full-frame imaging and zero-suppressed data, modification of the number of pixel data bits and independent programmability for distinct detector regions. In this work, a method is presented for seamlessly changing time frames and readout modes without data corruption while still ensuring that the data acquisition system (DAQ) does not need to stop and resynchronize at each change of setting, thus avoiding significant dead time. Data throughput is maximized by using a minimum unique data format, rather than lengthy frame headers, to differentiate between consecutive frames. A data control and transmitter (DCT) synchronizes data transfer from the pixel to the periphery, reconfigures the data to transmit it serially off-chip, while providing optimized decision support based on a DAQ definable mode. Measurements on a test structure demonstrate that the DCT can operate at 1 GHz in a 65 nm LP CMOS process. Full article

(This article belongs to the Special Issue Data Analysis for Smart Sensor Systems)

► Show Figures

Figure 1

17 pages, 3321 KiB

Open AccessArticle

Automatic Anomaly Detection on In-Production Manufacturing Machines Using Statistical Learning Methods

by Federico Pittino, Michael Puggl, Thomas Moldaschl and Christina Hirschl

Sensors 2020, 20(8), 2344; https://doi.org/10.3390/s20082344 - 20 Apr 2020

Cited by 41 | Viewed by 8771

Abstract

Anomaly detection is becoming increasingly important to enhance reliability and resiliency in the Industry 4.0 framework. In this work, we investigate different methods for anomaly detection on in-production manufacturing machines taking into account their variability, both in operation and in wear conditions. We [...] Read more.

Anomaly detection is becoming increasingly important to enhance reliability and resiliency in the Industry 4.0 framework. In this work, we investigate different methods for anomaly detection on in-production manufacturing machines taking into account their variability, both in operation and in wear conditions. We demonstrate how the nature of the available data, featuring any anomaly or not, is of importance for the algorithmic choice, discussing both statistical machine learning methods and control charts. We finally develop methods for automatic anomaly detection, which obtain a recall close to one on our data. Our developed methods are designed not to rely on a continuous recalibration and hand-tuning by the machine user, thereby allowing their deployment in an in-production environment robustly and efficiently. Full article

(This article belongs to the Special Issue Data Analysis for Smart Sensor Systems)

► Show Figures

Figure 1

15 pages, 6043 KiB

Open AccessArticle

Data Augmentation with Suboptimal Warping for Time-Series Classification

by Krzysztof Kamycki, Tomasz Kapuscinski and Mariusz Oszust

Sensors 2020, 20(1), 98; https://doi.org/10.3390/s20010098 - 23 Dec 2019

Cited by 45 | Viewed by 6790

Abstract

In this paper, a novel data augmentation method for time-series classification is proposed. In the introduced method, a new time-series is obtained in warped space between suboptimally aligned input examples of different lengths. Specifically, the alignment is carried out constraining the warping path [...] Read more.

In this paper, a novel data augmentation method for time-series classification is proposed. In the introduced method, a new time-series is obtained in warped space between suboptimally aligned input examples of different lengths. Specifically, the alignment is carried out constraining the warping path and reducing its flexibility. It is shown that the resultant synthetic time-series can form new class boundaries and enrich the training dataset. In this work, the comparative evaluation of the proposed augmentation method against related techniques on representative multivariate time-series datasets is presented. The performance of methods is examined using the nearest neighbor classifier with the dynamic time warping (NN-DTW), LogDet divergence-based metric learning with triplet constraints (LDMLT), and the recently introduced time-series cluster kernel (NN-TCK). The impact of the augmentation on the classification performance is investigated, taking into account entire datasets and cases with a small number of training examples. The extensive evaluation reveals that the introduced method outperforms related augmentation algorithms in terms of the obtained classification accuracy. Full article

(This article belongs to the Special Issue Data Analysis for Smart Sensor Systems)

► Show Figures

Figure 1

17 pages, 1683 KiB

Open AccessArticle

AI Radar Sensor: Creating Radar Depth Sounder Images Based on Generative Adversarial Network

by Maryam Rahnemoonfar, Jimmy Johnson and John Paden

Sensors 2019, 19(24), 5479; https://doi.org/10.3390/s19245479 - 12 Dec 2019

Cited by 15 | Viewed by 4601

Abstract

Significant resources have been spent in collecting and storing large and heterogeneous radar datasets during expensive Arctic and Antarctic fieldwork. The vast majority of data available is unlabeled, and the labeling process is both time-consuming and expensive. One possible alternative to the labeling [...] Read more.

Significant resources have been spent in collecting and storing large and heterogeneous radar datasets during expensive Arctic and Antarctic fieldwork. The vast majority of data available is unlabeled, and the labeling process is both time-consuming and expensive. One possible alternative to the labeling process is the use of synthetically generated data with artificial intelligence. Instead of labeling real images, we can generate synthetic data based on arbitrary labels. In this way, training data can be quickly augmented with additional images. In this research, we evaluated the performance of synthetically generated radar images based on modified cycle-consistent adversarial networks. We conducted several experiments to test the quality of the generated radar imagery. We also tested the quality of a state-of-the-art contour detection algorithm on synthetic data and different combinations of real and synthetic data. Our experiments show that synthetic radar images generated by generative adversarial network (GAN) can be used in combination with real images for data augmentation and training of deep neural networks. However, the synthetic images generated by GANs cannot be used solely for training a neural network (training on synthetic and testing on real) as they cannot simulate all of the radar characteristics such as noise or Doppler effects. To the best of our knowledge, this is the first work in creating radar sounder imagery based on generative adversarial network. Full article

(This article belongs to the Special Issue Data Analysis for Smart Sensor Systems)

► Show Figures

Figure 1

10 pages, 2898 KiB

Open AccessArticle

Data-Analytics Modeling of Electrical Impedance Measurements for Cell Culture Monitoring

by Elvira García, Pablo Pérez, Alberto Olmo, Roberto Díaz, Gloria Huertas and Alberto Yúfera

Sensors 2019, 19(21), 4639; https://doi.org/10.3390/s19214639 - 25 Oct 2019

Cited by 9 | Viewed by 4385

Abstract

High-throughput data analysis challenges in laboratory automation and lab-on-a-chip devices’ applications are continuously increasing. In cell culture monitoring, specifically, the electrical cell-substrate impedance sensing technique (ECIS), has been extensively used for a wide variety of applications. One of the main drawbacks of ECIS [...] Read more.

High-throughput data analysis challenges in laboratory automation and lab-on-a-chip devices’ applications are continuously increasing. In cell culture monitoring, specifically, the electrical cell-substrate impedance sensing technique (ECIS), has been extensively used for a wide variety of applications. One of the main drawbacks of ECIS is the need for implementing complex electrical models to decode the electrical performance of the full system composed by the electrodes, medium, and cells. In this work we present a new approach for the analysis of data and the prediction of a specific biological parameter, the fill-factor of a cell culture, based on a polynomial regression, data-analytic model. The method was successfully applied to a specific ECIS circuit and two different cell cultures, N2A (a mouse neuroblastoma cell line) and myoblasts. The data-analytic modeling approach can be used in the decoding of electrical impedance measurements of different cell lines, provided a representative volume of data from the cell culture growth is available, sorting out the difficulties traditionally found in the implementation of electrical models. This can be of particular importance for the design of control algorithms for cell cultures in tissue engineering protocols, and labs-on-a-chip and wearable devices applications. Full article

(This article belongs to the Special Issue Data Analysis for Smart Sensor Systems)

► Show Figures

Figure 1

Other

Jump to: Research

14 pages, 7559 KiB

Open AccessLetter

Smart Pack: Online Autonomous Object-Packing System Using RGB-D Sensor Data

by Young-Dae Hong, Young-Joo Kim and Ki-Baek Lee

Sensors 2020, 20(16), 4448; https://doi.org/10.3390/s20164448 - 9 Aug 2020

Cited by 17 | Viewed by 3690

Abstract

This paper proposes a novel online object-packing system which can measure the dimensions of every incoming object and calculate its desired position in a given container. Existing object-packing systems have the limitations of requiring the exact information of objects in advance or assuming [...] Read more.

This paper proposes a novel online object-packing system which can measure the dimensions of every incoming object and calculate its desired position in a given container. Existing object-packing systems have the limitations of requiring the exact information of objects in advance or assuming them as boxes. Thus, this paper is mainly focused on the following two points: (1) Real-time calculation of the dimensions and orientation of an object; (2) Online optimization of the object’s position in a container. The dimensions and orientation of the object are obtained using an RGB-D sensor when the object is picked by a manipulator and moved over a certain position. The optimal position of the object is calculated by recognizing the container’s available space using another RGB-D sensor and minimizing the cost function that is formulated by the available space information and the optimization criteria inspired by the way people place things. The experimental results show that the proposed system successfully places the incoming various shaped objects in their proper positions. Full article

(This article belongs to the Special Issue Data Analysis for Smart Sensor Systems)

► Show Figures

Figure 1

16 pages, 4066 KiB

Open AccessLetter

Quantization-Mitigation-Based Trajectory Control for Euler-Lagrange Systems with Unknown Actuator Dynamics

by Yi Lyu, Qiyu Yang and Patrik Kolaric

Sensors 2020, 20(14), 3974; https://doi.org/10.3390/s20143974 - 17 Jul 2020

Viewed by 1926

Abstract

In this paper, we investigate a trajectory control problem for Euler-Lagrange systems with unknown quantization on the actuator channel. To address such a challenge, we proposed a quantization-mitigation-based trajectory control method, wherein adaptive control is employed to handle the time-varying input coefficients. We [...] Read more.

In this paper, we investigate a trajectory control problem for Euler-Lagrange systems with unknown quantization on the actuator channel. To address such a challenge, we proposed a quantization-mitigation-based trajectory control method, wherein adaptive control is employed to handle the time-varying input coefficients. We allow the quantized signal to pass through unknown actuator dynamics, which results in the coupled actuator dynamics for Euler-Lagrange systems. It is seen that our method is capable of driving the states of networked Euler-Lagrange systems to the desired ones via Lyapunov’s direct method. In addition, the effectiveness and advantage of our method are validated with a comparison to the existing controller. Full article

(This article belongs to the Special Issue Data Analysis for Smart Sensor Systems)

► Show Figures

Figure 1

Journal Menu

Journal Browser

Data Analysis for Smart Sensor Systems

Share This Special Issue

Special Issue Editor

Special Issue Information

Keywords

Benefits of Publishing in a Special Issue

Published Papers (12 papers)

Research

Other

Further Information

Guidelines

MDPI Initiatives

Follow MDPI