Recognition of moving ground targets by measuring and processing seismic signal
Introduction
There are two ways of detecting the moving ground vehicle targets. One way is to send a signal to the environment and measure the interaction between this signal and the environment, and target recognition is achieved by analyzing the reflection wave. This is called active detection technology. Examples include sonar and radar. Due to large volume and wave launch, this kind of detection equipment discloses itself easily if it is used in battlefield surveillance. Furthermore, they are expensive because of much extra hardware. The other alternative, called passive detection technology where passive sensors simply record signals already present in the environment and utilize some signals that are produced by targets. These signals include seismic, acoustic, infrared signals and so on [1], [2], [3]. Because the passive detection equipment need not emit any signal to the environment, it is difficult to be found. At the same time, the passive detection equipment is cheaper and smaller than the active ones because they do not need lot of additional hardware.
For the passive detection methods, most of the target detection and classification methods have been proposed based on processing of acoustic signal, image and infrared signals because seismic waves are more complicated to be analysed [4], [5], [6]. Seismic waves propagate in different forms, different directions, different speeds, and are highly dependent on the underlying geology. So, up to now, seismic signals usually are used in target trace instead of target recognition.
However, seismic sensors also have advantages over other systems. Vehicle recognition by means of sound, image and infrared signals will be affected by Doppler effects, by noises introduced from various moving parts of vehicles, and by atmospheric and terrain variations, while seismic waves are less sensitive to these factors. Seismic sensors provide nonline of sight detection capabilities for vehicles at significant ranges. Table 1 shows a rough outline of seismic sensing capabilities for a variety of vehicle targets. If seismic signal and acoustic signal are all used in battlefield surveillance, the seismic sensors can also provide early detection of many targets and process unique detection capabilities in the case of adverse acoustic propagation conditions [7].
On the other hand, future development promises low-cost, robust systems of small size that can be deployed by many means. The requirements for some fields, especially in the battlefield surveillance application are that these sensors are cheap, can be fitted into a small volume and their power consumption must be suitable for battery operated devices. In the past, high performance sensor payloads were too big and expensive and easy to be lost on the battlefield. However, recent technical breakthroughs in micro sensor work are making payload and employment concepts operationally affordable. Micro-electro-mechanical system (MEMS) devices can meet these requirements since they can be batch-fabricated and similar advantages as for standard integrated circuits are envisaged. MEMS sensors are very cheap, and have small volume, light weight and low power [8], [9], [10]. In particular, the seismic detection method is easy to be developed by MEMS technology. Because of the advantages of the seismic detection method mentioned above, recognition of moving ground targets by measuring and processing the seismic signal is discussed in the paper.
Because seismic feature extraction of vehicle targets is an important index of target recognition [11], [12], the seismic features of typical vehicles have been tested and analyzed in this paper. In order to realize target classification and recognition of a single sensor, based on the multi-layer feedforward neural networks and its training, the ANN data fusion structure of seismic signals for target classification and recognition is discussed. The BP algorithm is used in multi-layer feedforward neural networks. Because its training speed is not rapid enough, a new learning algorithm on the basis of traditional BP algorithms is put forward. Comparing the experiment results, it can be confirmed that the improved BP is effective in classification and recognition of seismic signals for vehicle targets.
Section snippets
Seismic properties’ test and result of vehicle targets
Tests needed to be done outdoors so as to get seismic properties of three different targets. It lays a foundation for target classification. The test was done on provided ground. The targets included diesel engine vehicle, heavy diesel engine vehicle and gasoline engine vehicle. It was a sunny day, 15–20 °C temperature, and 2–3 wind-force levels.
First, the seismic signals from a vehicle target were acquired by seismic sensors. Second, the signals were stored in a large dynamic range, low noise,
Mechanism of vehicle target seismic signal
Vehicles can be detected by a seismic velocity transducer. Vehicles moving over ground generate a succession of impacts; these soil disturbances propagate away from the source as seismic waves. For ground vehicles, the wave of importance is the Rayleigh wave. This wave is generated by the impact of vehicles on the ground surface, the broadband component comes from surface irregularities, the narrow band component from the periodic impact of the vehicles.
The seismic signal is varied to different
Determining data fusion method and architecture of the seismic target recognition
Data fusion is the process of combining information from a lot of data source to produce a unified result. For systems with sensors, the sensing process generally follows the pattern: detection, preprocessing, fusion, and data interpretation. Each step depends on the results of the immediately preceding step. The raw data (detection) is put into some sort of computable format (preprocessing). The computable data from source is combined (fusion) and evaluated by the system (interpretation). The
Application result
Improved BP algorithm data fusion is used in target classification and recognition for actual measurement seismic data. The result of improved BP algorithm data fusion to target classification and recognition is as shown in Table 2. Row “Total number of sample” represents actual samples, “Correct recognition number of sample” and “Incorrect recognition number of sample”, respectively, represent how many samples are correctly and falsely recognized. “Recognition rate” represents correct
Conclusion and discussion
In this paper, we have successfully applied neural networks data fusion to recognition of seismic signals for vehicle targets. In order to use neural networks more efficiently, the learning method is improved. According to experiments, seismic signals can be used to detect vehicles using the surface Rayleigh waves they generate, target seismic properties acquired outdoor are correct. And neural networks data fusion to target classification and recognition can acquire a high recognition rate.
Jinhui Lan obtained her PhD degree from Beijing Institute of Technology, China in 1998. Since then, Dr. Jinhui Lan was a lecturer at Tsinghua University, China, where she taught and researched in measurement and instrument, data fusion, sensor, MEMS technology, etc. From August, 2002 to February, 2004, she was engaged in multi-sensor data fusion research at Deakin University, Australia, as a visiting academic. Since July, 2004, she became an associate professor at University of Science and
References (23)
- D. Estrin, L. Girod, G. Pottie, M. Srivastava. Instrumenting the world with wireless sensor network, in: Proceedings of...
- et al.
Wavelet-based ground vehicle recognition using acoustic signal
Wavelet Applications 3
(1996) - et al.
An investigation of acoustic-to-seismic coupling to detect buried antitank landmines
IEEE Transactions on Geoscience and Remote Sensing
(2001) - et al.
Acoustic target tracking and target identification-recent results
Unattended Ground Sensor Technologies and Applications, SPIE
(1999) - et al.
Acoustic target detection, tracking, classification, and location in a multiple target environment
Peace and Wartime Applications and Technical Issues for Unattended Ground Sensors SPIE
(1997) - et al.
Problems in seismic detection and tracking
Unattended Ground Sensor Technologies and Applications II
SPIE
(2000) - L.B. Stotts, Unattended ground sensor related technologies: an army perspective. In Unattended Groung Sensor...
- et al.
A batch-fabricated silicon accelerometer
IEEE Transactions Electron Devices
(1979) - et al.
Surface micromachined accelerometers
IEEE J. Solid-State Circuits
(1996) - et al.
Fundamentals of Microelectromechanical Systems
(2001)
Multisensor maximum attribute data fusion method and its application in target classification
Journal of Tsinghua University (Science and Technology)
Cited by (36)
Moving target recognition with seismic sensing: A review
2021, Measurement: Journal of the International Measurement ConfederationCitation Excerpt :They altered the ANN and designed a backpropagation neural network, which improved the classification accuracy on the basis of the naive ANN. In their follow-up work [108], the three vehicles were characterized by differences in the frequency and magnitude of the peaks in the spectrum. The proposed ANN could improve the learning speed and avoid local minimum values in the error curve.
Training deep neural networks for wireless sensor networks using loosely and weakly labeled images
2021, NeurocomputingCitation Excerpt :Wireless sensor networks (WSNs) typically are designed to detect and identify neighboring objects in wild [1–4] with sound or vibration sensors in the form of single [5] or microarrays [6].
A quarter-car vehicle model based feature for wheeled and tracked vehicles classification
2013, Journal of Sound and VibrationCitation Excerpt :Currently, the features employed by researchers for ground target classification are normally extracted from spectra or time domain of seismic signals. Zhang [5] and Lan [6,7] used the structure and the amplitude of frequency spectrum as the features which are calculated by Shot-time Fourier transform (STFT). A technique of Time Encoded Signal Processing and Recognition (TESPAR) [8] for vehicles classification was introduced which can extract features in time domain effectively.
Target Classification in Unattended Ground Sensors With a Two-Stream Convolutional Network
2023, IEEE Sensors JournalPattern-of-Life Activity Recognition In Seismic Data
2022, Applied Artificial Intelligence
Jinhui Lan obtained her PhD degree from Beijing Institute of Technology, China in 1998. Since then, Dr. Jinhui Lan was a lecturer at Tsinghua University, China, where she taught and researched in measurement and instrument, data fusion, sensor, MEMS technology, etc. From August, 2002 to February, 2004, she was engaged in multi-sensor data fusion research at Deakin University, Australia, as a visiting academic. Since July, 2004, she became an associate professor at University of Science and Technology Beijing, China. In cooperation with other scholars, she has finished more than 10 projects and succeeded in solving some key technical problems. So far, she has published over 30 refereed papers in the areas of measurement and instrument, multi-sensor data fusion, and MEMS technology.
Saeid Nahavandi received a B.Sc. (Hons.), M.Sc., and Ph.D. in automation from Durham University, UK. In 1991 he joined Massey University, New Zealand, where he taught and led research in robotics. In 1998 he joined Deakin University and now holds the Chair in Engineering. Professor Nahavandi is the leader of the Intelligent Systems Research Group and also Manager for the Cooperative Research Centre for CAST Metals Manufacturing. He has published over 150 reviewed papers, delivered several invited plenary lectures at international conferences, and is the recipient of four international awards in engineering. Dr. Nahavandi is the founder of the World Manufacturing Congress series and the Autonomous Intelligent Systems Congress series and the Editor (for South Pacific) of Intelligent Automation and Soft computing International Journal. He is a Fellow of IEAust, IEE and the member of IEEE.
Tian Lan, PhD student in OGI school of Science and Engineering of OHSU, USA. Major in Electrical and Computer Engineering. Research interests include: Signal and Image Processing, Machine Learning, Neural Networks, Sensori-motor Integration and Control.
Yixin Yin received Bachelor Degree, Master Degree, and Ph.D. in automation from University of Science and Technology Beijing (USTB), China. Professor Yixin Yin is the head of Information Engineering School, USTB. He has published over 40 reviewed papers, obtained four national awards in automation. Dr. Yixin Yin is one of leaders of Chinese Association of Automation.