Magnetic gradient full-tensor fingerprints for metallic objects detection of a security system based on anisotropic magnetoresistance sensor arrays

Because of the abuse of dangerous metallic objects such as knives, guns, etc., numerous public security incidents come out thick and fast in many countries.^1,2 A commonly used and accepted commercial method for security detection is the electromagnetic (EM) based approach, in which an induction coil is adopted to sense suspicious metallic objects, such as hand-held metal panel detectors,³ walk through metal detectors,⁴ etc.^5,6 To be specific, the security door can detect the covert carried metallic objects because the two side doors are respectively equipped with an induction coil for receiving and transmitting alternating EM signals.⁷ In case the metallic objects are polarized by the alternating EM signal, an eddy current would be generated, and thus a secondary magnetic field with the same magnitude in the direction opposite to the initial magnetic field could be produced.⁸ The metallic object can be detected by measuring the eddy current signal. However, due to the limitation of the EM working principle, this kind of system is always heavy, requires high power, and is bulky in size,⁹ except the mine-detectors which are mainly employed in military applications. There are also other kinds of EM-based approaches including X-ray,¹⁰ terahertz compressive imaging,¹¹ millimeter wave imaging,¹² etc.^13,14 Likewise, these methods still have the aforementioned shortcomings, making them rarely used now.

Generally, the time-dependent geomagnetic field should be mutated by metallic materials because of their relatively higher magnetic permeability.^15,16 Such a kind of minute mutation in the earth’s magnetic field can be detected by a sensitive magnetic sensor, dubbed weak magnetic detection (WMD).^17,18 The main difference between the EM and WMD is that the WMD operates in a passive way, which is initially adopted in military investigations.¹⁹ Up to now, the WMD method has been employed in many applications such as resource prospection,²⁰ unexploded ordnance detection,²¹ etc. Moreover, the magnetic characteristics of ocean mines,²² unmanned aerial vehicles,²³ and brain imaging²⁴ have been extensively investigated in recent studies, which can be used to describe the temporal and unique distribution of the targets’ magnetic field. The fact that most of the dangerous objects are made of metallic materials makes WMD a promising detection method for public security system.²⁵ However, to the best of our knowledge, there is no other work which has investigated the feasibility of WMD based magnetic imaging for public security applications, especially the magnetic gradient full-tensor fingerprint imaging based approach.

One of the many challenges in designing a WMD based public security system may lie in the difficulty to distinguish the weak object signals (i.e., scissors and hammers) from unknown interference, which would decrease the signal-to-noise ratio (SNR) and the range of the detection zone.²⁶ In this study, aiming to overcome the aforementioned problems, the possibility of detecting such magnetic anomalies using a microelectromechanical system (MEMS) anisotropic magnetoresistance (AMR) sensor is investigated. A magnetic gradient full-tensor configuration arranged on a planar cross structure was employed, which can further suppress the background noise and reduce the orientation and orthogonality errors. The contributions of this work include the following points:

1. A new magnetic gradient full-tensor fingerprint based security system is developed, which integrates three sets of magnetic sensor arrays, a microcontroller, a battery, and a personal computer (PC).

2. An efficient magnetic anomaly data processing framework is implemented, which consists of data acquisition, data processing, noise reduction, and magnetic fingerprinting.

3. Our proposed security system can effectively configure metallic objects more than 50cm and identify multiple objects separated by less than 20 cm.

The remainder of this paper is arranged as follows: Section II explains the preliminary theory of the magnetic gradient tensor and the characterization of the AMR magnetic sensor. Section III considers the development of the security system and the operation principle. Numerous validation results presented in Sec. IV indicate the superiority of the designed security system, and a case of the appropriate object feature segmentation method is discussed. Finally, this paper is summarized in Sec. V, highlighting contributions and future expectations.

In a rectangular coordinate system o-xyz, the magnetic strength vector B can be expressed by its three orthogonal components as (B_x, B_y, B_z). Likewise, the magnetic gradient tensor G can be derived through calculating the rate-of-change along each direction of (B_x, B_y, B_z) as²⁷ 

Generally, G is both symmetric and traceless in free space, and thus there is B_xy = B_yx, B_xz = B_zx, B_yz = B_zy, and B_zz = −B_xx − B_yy. Hence, Eq. (1) can be rewritten as

From Eq. (2), it can be seen that there are five independent elements among the nine elements of G. Hence, G is a real symmetric matrix, and three invariants of the magnetic gradient tensor can be derived directly as

Through measuring the five independent elements, we can obtain all the nine elements of G, and thus the metallic object could be determined. The greatest advantage of using a magnetic gradient tensor is that it is entirely passive when compared with traditional active electromagnetic detection, which can guarantee a long working time. Furthermore, we can obtain nine different magnetic fingerprints to enlarge the effective information of the magnetic object, further improving the detection performance.

For the possibility of detecting such metallic objects, we set up an improved AMR magnetic sensor, dubbed QMC5883L. The operating principles of an AMR magnetometer are significantly different from those of a vector magnetometer, e.g., Hall magnetometers, fluxgate, etc. The QMC5883L module is an improved version of a magneto-resistive sensor embedded with an application specific integrated circuit (ASIC) containing amplification, automatic degaussing strap drivers, offset cancellation, temperature compensation, and a 16-bit analog-to-digital converter (ADC). After initial tests with this sensor, it achieves a field resolution of 1 mG in ±8 G fields and a low power consumption of 75 μA at a low voltage operation from 2.16 V to 3.6 V. In this paper, for the parameters of the sensor, the measurement range is ±0.88 G, the gain is 1370 LSB/G, and the output data rate is 0.75 Hz.

Figure 1 illustrates the internal structure of the QMC5883L module. The controlling of the set or reset function is done automatically by the ASIC for each measurement. One half of the difference from the measurements taken after a set pulse and after a reset pulse will be put in the data output register for each of the three axes. By doing so, the sensor’s internal offset and its temperature dependence are removed or canceled for all measurements.

Generally, the separation of or spacing between the arrays has a large impact on the overall design of the security system. The smaller the separation, the greater the required number of sensors, which will increase the cost and complexity. In this study, we trialed numerous sensor separations, and after analyzing the results, four AMR sensors arranged on a planar cross structure with 20 cm spacing were implemented, as shown in Fig. 2, which was found to be a good compromise between system complexity and spatial resolution. In this case, although the selected sensor separation d means that the vertical or horizontal resolution can only be guaranteed as less than 20 cm, it is not particularly useful in object analysis and characterization of the magnetic gradient full-tensor fingerprints.

For the structure of the sensor array, the magnetic gradient full-tensor G can be derived based on Eq. (2) as

where m in B_mn stands for the number of each sensor, n stands for the magnetic intensity along x, y, z directions.

The main connection and block diagram of the security system is shown in Fig. 3. The upper part is duplicated three times to make 12 channels in total. The magnetic anomaly signals captured from the QMC5883L sensor are transferred to the data register in the microcontroller. Aiming to improve the SNR of the collected data, a data filtering module using least squares smoothing was used to suppress the unexpected data fluctuations. Hence, the processed data was sent to the PC for magnetic gradient full-tensor fingerprinting.

Figure 4 shows the cardinal direction of the sensor array and the overall security system establishment. The array is aligned with the central axis of the detector panel to pick up any magnetic anomalies within the panel width due to the presence of metallic objects. If no anomaly is present in magnetic gradient full-tensor fingerprints, the magnetic field collected by the sensor array is unchanged. Likewise, the existence of a metallic object will cause a distortion in the magnetic field, which can thus be detected by the sensor array.

In this section, the detection distance of the proposed security was evaluated. Several dangerous objects including a mobile phone, a chopping knife, a hammer, and a kitchen knife were employed as the targets, which were invisibly placed in a pocket of a human being. Figure 5 shows the overall practical AMR based security system which consists of three sensor arrays, a microcontroller, a communication module, and a PC. The speed of the walking pass holder was about 1 m/s, which was consistent with the actual situation. Generally, the magnetic field strength will drop rapidly as the distance increases. Hence, in this case, the distance between the target and the sensor array was tested from near to far until the distance was not detectable.

Table I shows the approximate maximum detectable distance of 60 cm, 70 cm, 80 cm, and 50 cm with respect to the mobile phone, chopping knife, hammer, and kitchen knife, respectively. We observe that the proposed security system is relatively sensitive and can detect all the aforementioned small targets which are no less than 50 cm, which demonstrates that the proposed magnetic gradient full-tensor based approach can offer a viable strategy for public security system applications.

As mentioned, the collected magnetic anomaly data can be transferred from the security system to the PC, where the magnetic gradient full-tensor fingerprint can be yielded. Figure 6 shows one magnetic fingerprinting result of the three sensor arrays for each target when the detected distance is 50 cm. It can be seen that the fluctuations of the magnetic gradient full-tensor element B_yy are conducted as a function of sampling time. The middle sensor array module shows the strongest response because it has the smallest distance with the targets. Furthermore, we observe that different targets have their own unique identifiable magnetic patterns to be discriminated, which demonstrates that the magnetic gradient tensor fingerprint is an effective identification feature for metallic object detection.

Figure 7 presents the nine elements’ variation curvature of the magnetic gradient full-tensor G [refer to Eq. (4)] for the kitchen knife through mapping out the magnetic anomalies from the middle sensor array module. It can be seen that when the kitchen knife holder passes through the security system, different magnetic anomalies are obtained respectively in each of the magnetic gradient full-tensor elements between a sampling time of 2 s to 3 s. The results further demonstrate that the nine independent elements can manifest unique identifiable magnetic patterns for different metallic objects.

The invisible carry-on dangerous metallic objects will not always be independent, i.e., sometimes they could be placed near each other. Hence, the ability of the security system to identify multiple objects is an important character. To evaluate this capability, the hammer and mobile phone were placed on the front and back of the walking pass holder separated by 20 cm. In this case, the detection distance between the two targets and the sensor array is set to be 50 cm, and the walking pass speed is also about 1 m/s.

Figure 8 presents the nine elements’ variation curvature of the magnetic gradient full-tensor G for the kitchen knife and the mobile phone through mapping out the magnetic anomalies from the middle sensor array module. It can be seen that when compared with the results of one target, as shown in Fig. 7, the number of the magnetic anomaly fluctuations, as well as the corresponding magnitude, is increased. This phenomenon should be caused by the superposition or mutual perturbation of the magnetic field generated by the two objects.

Figure 9 shows one magnetic gradient full-tensor fingerprint B_yy of the two objects. It can be seen that two magnetic anomalies with respect to the hammer and the mobile phone are obviously detected by the sensor array, respectively. From Table I, we perceive that the hammer has a longer detection distance than the mobile phone, which implies that the hammer should generate stronger magnetic field intensity with the same distance when compared with the mobile phone. The results illustrated in Fig. 9 are consistent with this speculation, where the magnetic gradient tensor fingerprint strength of the hammer is stronger than that of the mobile phone. Consequently, the AMR based security system shows great potential in multiple objects’ feature segmentation without losing their own characteristics of the magnetic gradient full-tensor fingerprints.

According to the aforementioned investigations for different magnetic anomalies, a LabVIEW monitoring interface, as shown in Fig. 10, was programmed to monitor the detection results when the magnetic data were transferred from the acquisition system to the PC. The time-varying magnetic data collected by the sensor arrays can be monitored by the program, and the abnormal alarm of the metallic target holder can be shown simultaneously by one lamp “alarm light” in the upper-left part. In this case, when this lamp turns red, there should be a suspicious target holder passing by. Otherwise, this lamp should keep itself in green. Furthermore, the x, y, z axial magnetic field of each sensor is real-time displayed in the front panel, and the corresponding magnetic gradient full-tensor fingerprints were employed to further make an accurate pass decision.

In this study, a new magnetic gradient full-tensor fingerprint based security system using AMR magnetic sensor arrays was developed. System design, data acquisition, and real-time magnetic fingerprint imaging were also investigated. Commonly used metallic targets, including mobile phone, hammer, etc., were adopted as samples to validate the system and as a verification of the proposed concept. Furthermore, the pattern extraction and feature segmentation were implemented to evaluate the capacity of discrimination between different metallic targets using the proposed security system. Consequently, the proposed new security system based on magnetic gradient full-tensor fingerprint shows a great scope for metallic target detection and discrimination. In addition, the system’s two foremost advantages are (1) the nine elements of the magnetic gradient full-tensor can supply more reference information, further suppressing the background noise and reducing the orientation and orthogonality errors of the magnetic sensor and (2) the WMD based technology allows the system to work in a passive way, making it a promising candidate to be employed in unconstrained environments, and overcome the shortcomings of the EM based system such as heaviness, high power, bulky size, etc.

We have demonstrated the feasibility of the magnetic gradient full-tensor fingerprint based approach used in the security system. However, due to the own characteristics of the AMR magnetic sensor, the magnetic field measurement accuracy is relatively low which would influence the detectable distance. Hence, our future work will focus on investigating a calibration or compensation method for this kind of sensor data and attempt to use a new MEMS magnetic sensor RM3100, to further improve the performance of the proposed security system.

This work was partly supported by the National Natural Science Foundation of China under Grant Nos. 41904164 and 41874212, the Foundation of the National Key R&D Program of China under Grant No. 2018YFC1503702, the Foundation of Science and Technology on Near-Surface Detection Laboratory under Grant Nos. 6142414180913, TCGZ2016A005, and 614241409041217, and the Fundamental Research Funds for the Central Universities, China University of Geosciences (Wuhan), under Grant No. CUG190628.