IoT-based critical infrastructure enabled radar information fusion

https://doi.org/10.1016/j.compeleceng.2022.107723Get rights and content

Abstract

With the development of Internet of things (IoT) and critical infrastructure in modern industry, many technologies are facing serious challenges brought by the change of application. For instance, the real-time radar signal processing is based on the IoT and infrastructure. In particular, parameter estimation based on Radar Cross Section (RCS) from narrow-band radar requires sufficient data and effective algorithm. In this paper, the geometric parameters of the target are estimated by the Deep Learning method. Moreover, considering the difference of the information collected by different radars, a data fusion method based on channel attention mechanism is proposed for multi-site RCS time-series, which greatly improves the performance of parameters estimation.

Introduction

The development of Internet of things (IoT) and critical infrastructure in modern industry give a splendid opportunity for new application of advanced technologies to flourish. Radar, as a perception sensor, undertakes some sensing functions in IoT system [1], [2]. Target perception and analysis based on radar signal plays an important role in military, aerospace, transportation, and many other fields [3], [4]. Along with the progress of space technology, exploring a low-cost, high-precision, fast and convenient space object perception solution has become the key point of this field.

The space target can be identified based on its shape information and Micro-motion of mid-course, while the estimation of its specific size requires tremendous features in radar information, such as High Resolution Range Profile (HRRP) and Synthetic Aperture Radar (SAR) Imagery. Therefore, few researchers utilize RCS data to estimate target size. In Ref. [5], the target is equivalent to an ellipsoid, and then its major and minor axes are estimated. On the basis of this research, Ref. [6] proposed a method to estimate the ratio of major and minor axes without modeling. A target size estimation method based on RCS sequence variance is proposed in Ref. [7]. In Ref. [8], a new geometric parameter estimation curve was established by using RCS acquired during a measurement campaign in an anechoic chamber.

Narrow-band radar which we use in this paper, has the advantages of low cost and the high rate of the equipment. The RCS of the target from the narrow-band radar is a physical quantity to quantize the scattering intensity of a target. However, compared with information of the signal from wide-band radar, the RCS contains less information so that needs sufficient data and an effective algorithm to achieve the effective estimation of the target’s size.

Nowadays, Neural Network has been widely used in a series of information processing technology fields [9], [10], [11], [12], especially in the field of Computer Vision and Natural Language Processing. A series of research on radar target recognition based on the Deep Learning has been achieved continuously. Ref. [13] presents a Deep Learning method with Recurrent neural Network (RNN) for HRRP feature recognition. In Ref. [14], the samples are divided into frames according to the azimuth of target’s radar signal, and then the shared features between frames are extracted by using Deep Belief Network. Ref. [15] uses the Deep Learning algorithm based on Convolutional Neural Network (CNN) to recognize military vehicles from the 2D images collected by SAR, and obtains good recognition results. The HRRP data of three kinds of aircraft are classified by using the RNN with attention mechanism in Ref. [16]. In Ref. [17], the Deep Learning method based on radar signal time–frequency map and time-domain information fusion is used for Micro-motion state classification and target identification. In Ref. [18], the Micro-motion of target is used to identify the warhead and decoy based on CNN.

In fact, when we describe one target using various radar signals from different positions, there exists gaps due to the difference of relative attitude. And these different complementary radar signals should to be well utilized. Using the method based on data fusion will be an effective way to make use of these complementary signals, and an elaborate fusion method design is a good way to improve the efficiency of data and information fusion.

Improvement of IoT brings the connection between industrial facilities becomes more efficient and speedy, which makes it more feasible to process the information obtained from different geographical locations at the same time. This paper explores radar perception patterns under the critical Infrastructure of modern industry and proposes a multi-site radar data fusion network based on Deep Learning. From the perspective of neural network architecture and the sequential characteristic of data, the structure of the main body of the network is designed, the information from the three radar sites is fused at the same time, and the better of parameter estimation is obtained. What is more, the attention mechanism is introduced into our fusion algorithm, and a more efficient data fusion module is designed, which further improves the capability of our model.

The main contribution of this paper is listed as follows:

1. Considering the current situation of radar target size parameter estimation, this paper applies the Deep Learning method to target size parameter estimation of RCS time series. The method’s effectiveness is explored from several aspects: the radar position, time sequence length, and model structure.

2. Considering the complementarity of radar signals from different radars observing the same target, this paper proposes a data fusion method based on Channel attention mechanism, which can gate the information from different radars from the channel level to achieve effective data fusion.

3. In this experiment, the target size parameter estimation effect based on a single radar reflects various characteristics in different aspects. The data fusion method we proposed can effectively improve the effect of parameter estimation. Compared with the experimental results based on single radar information, the RMSE of the cone height parameter is reduced by 45.1541 mm on average. The RMSE of the half cone angle parameter is reduced by 0.2345°on average. Simultaneously, compared with the data fusion method based on simple linear mapping, the two indexes are reduced by 15.0658 mm and 0.065°respectively, which shows the high efficiency of the method in data fusion.

The remaining part of the paper has been arranged as follows. Section 2 presents investigation methods with relevant theoretical background. Method details are described in Section 3. Experimental details and the results of the present investigation on parameters estimation are described in Section 4. The results are critically discussed in Section 5. Finally, key conclusions are presented in Section 6.

Section snippets

Influence of geometry parameters on RCS

The target’s geometric shape is one of the sources of RCS attitude sensitivity, so the change of geometric shape parameters will affect the RCS information distribution. The RCS data based on narrow-band radar has limited ability to describe the target, and the signal is expressed in the form of sub-scattering points, which can only be analyzed through intensity information and fluctuation characteristics. Therefore few researchers estimate size parameter estimation was carried out based on RCS

RCS turntable data simulation

Samples are the basis of data analysis, especially for intelligent algorithms like Deep Learning. Due to the consideration of security and national defense secrecy, both the measured radar signal time series and the omni-directional radar imaging information of the target itself are non-public confidential information. It is difficult for us to obtain the radar imaging information of the target. Therefore, the data based on simulation is the basis of the feasibility of this work. In this paper,

Experiments

RCS time series contains a lot of target information, so it is feasible to use Deep Learning model to estimate the parameters of target geometry information. Therefore, we use a variety of network models to estimate the parameters of the target RCS time series. The experiment in this paper is based on the data simulation and generation method introduced in the previous paper.

Considering the essence of the time series, different length time series may contain different level of size information.

Discussion

Due to the low information amount in the RCS time series, little researcher estimate target size based on RCS time series. Considering the excellent performance of Deep Learning methods in data processing currently, this paper used the Deep Learning model to estimate the size parameters based on RCS time series for the first time. The feasibility of target size estimation using RCS data has been evaluated.

And this paper analyzes the target size parameter estimation task from the radar position,

Conclusion

In this paper, we explore the Deep Learning’s ability to estimate the geometric parameters of the target, which can be discussed from three aspects of the network model. They are radar position, time sequence length, and model structure. The phenomena are obvious. The data collected at different radar positions affect the parameter estimation results. Among our data, the experimental result based on the data collected by radar 1 is the best, while the based on radar 3 is the worst; for the

CRediT authorship contribution statement

Jiachen Yang: Conceptualization, Supervision, Resources, Project administration. Zhuo Zhang: Methodology, Software, Data curation, Writing – original draft. Wei Mao: Software, Methodology. Yiwen Sun: Data curation, Writing – review & editing. Yongjun Bao: Supervision, Resources. Houbing Song: Conceptualization, Project administration.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This work was partially supported by National Natural Science Foundation of China (No. 61871283), the Foundation of Pre-Research on Equipment of China (No. 61400010304) and Major Civil-Military Integration Project in Tianjin, China (No. 18ZXJMTG00170).

Jiachen Yang received the M.S. and Ph.D. degrees in communication and information engineering from Tianjin University, China, in 2005 and 2009, respectively. He is currently a professor at Tianjin University. His research interests include stereo vision research, pattern recognition and image quality evaluation.

References (30)

  • Ghio S, Martorella M. Size estimation of space debris models from their RCS measured in anechoic chamber. In: 2020 17th...
  • He K, Zhang X, Ren S, et al. Deep residual learning for image recognition. In: Proceedings of the IEEE conference on...
  • VaswaniA. et al.

    Attention is all you need

  • FawazH. et al.

    Deep learning for time series classification: a review

    Data Min Knowl Discov

    (2019)
  • YangJ. et al.

    Blockchain-based sharing and tamper-proof framework of big data networking

    IEEE Netw

    (2020)
  • Cited by (5)

    Jiachen Yang received the M.S. and Ph.D. degrees in communication and information engineering from Tianjin University, China, in 2005 and 2009, respectively. He is currently a professor at Tianjin University. His research interests include stereo vision research, pattern recognition and image quality evaluation.

    Zhuo Zhang received the M.S. degree in electronic information engineering from Tianjin University, Tianjin, China, in 2021. He is currently pursuing the Ph.D. degree at the school of electrical and information engineering, Tianjin University, Tianjin, China. His research interests include radar signal processing and Deep Learning.

    Wei Mao received the M.S. degree in Aircraft design from China Academy of Launch Vehicle Technology, China, in 2014. He is currently a engineer at China Academy of Launch Vehicle Technology. His research interests include Space target recognition, pattern recognition and computer simulation.

    Yiwen Sun is currently pursuing the M.S. degree with the School of Electrical and Information Engineering, Tianjin University, Tianjin, China. His research interests include generalization problem in reinforcement learning, time series processing, neural architecture search and edge computing.

    Yongjun Bao received the M.S. degree from the Peking University, Beijing in 2007. He is the senior director of Technology and Data Center, JD.com. He leads large-scale machine learning framework, knowledge graph and online advertising etc. He was the winner of scene parsing in places challenge 2017. His current research interests include machine learning, computer vision and computational advertising.

    Houbing Song received the Ph.D. degree in electrical engineering from the University of Virginia, Charlottesville, VA, in 2012. He is currently an Assistant Professor and the Director of the Security and Optimization for Networked Globe Laboratory. His research interests include cyber–physical systems, cybersecurity and privacy, internet of things, edge computing, AI/machine learning, big data analytics, unmanned aircraft systems, connected vehicle.

    This paper is for special section VSI-cei. Reviews were processed by Guest Editor Dr. Imran Razzak and recommended for publication.

    View full text