Machine Learning Based Localization and Classification with Atomic Magnetometers

Automated Machine Learning Strategies for Multi-Parameter Optimisation of a Caesium-Based Portable Zero-Field Magnetometer

Machine learning (ML) is an effective tool to interrogate complex systems to find optimal parameters more efficiently than through manual methods. This efficiency is particularly important for systems with complex dynamics between multiple parameters and a subsequent high number of parameter configurations, where an exhaustive optimisation search would be impractical. Here we present a number of automated machine learning strategies utilised for optimisation of a single-beam caesium (Cs) spin exchange relaxation free (SERF) optically pumped magnetometer (OPM). The sensitivity of the OPM (T/Hz), is optimised through direct measurement of the noise floor, and indirectly through measurement of the on-resonance demodulated gradient (mV/nT) of the zero-field resonance. Both methods provide a viable strategy for the optimisation of sensitivity through effective control of the OPM's operational parameters. Ultimately, this machine learning approach increased the optimal sensitivity from 500 fT/Hz to <109fT/Hz. The flexibility and efficiency of the ML approaches can be utilised to benchmark SERF OPM sensor hardware improvements, such as cell geometry, alkali species and sensor topologies.

Light scattering control with the two-step focusing method based on neural networks and multi-pixel coding

Focusing light through scattering media is essential for high-resolution optical imaging and deep penetration. Here, a two-step focusing method based on neural networks (NNs) and multi-pixel coding is proposed to achieve high-quality focusing with theoretical maximum enhancement. In the first step, a single-layer neural network (SLNN) is used to obtain the initial mask, which can be used to focus with a moderate enhancement. In the second step, we use multi-pixel coding to encode the initial mask. The coded masks and their corresponding speckle patterns are used to train another SLNN to get the final mask and achieve high-quality focusing. In this experiment, for a mask of 16 × 16 modulation units, in the case of using 8 pixels in a modulation unit, focus with the enhancement of 40.3 (only 0.44 less than the theoretical value) has been achieved with 3000 pictures (1000 pictures in the first step and 2000 pictures in the second step). Compared with the case of employing only the initial mask and the direct multi-pixel encoded mask, the enhancement is increased by 220% and 24%. The proposed method provides a new idea for improving the focusing effect through the scattering media using NNs.

Detection and Characterisation of Conductive Objects Using Electromagnetic Induction and a Fluxgate Magnetometer

Eddy currents induced in electrically conductive objects can be used to locate metallic objects as well as to assess the properties of materials non-destructively without physical contact. This technique is useful for material identification, such as measuring conductivity and for discriminating whether a sample is magnetic or non-magnetic. In this study, we carried out experiments and numerical simulations for the evaluation of conductive objects. We investigated the frequency dependence of the secondary magnetic field generated by induced eddy currents when a conductive object is placed in a primary oscillating magnetic field. According to electromagnetic theory, conductive objects have different responses at different frequencies. Using a table-top setup consisting of a fluxgate magnetometer and a primary coil generating a magnetic field with frequency up to 1 kHz, we were able to detect aluminium and steel cylinders using the principle of electromagnetic induction. The experimental results were compared to numerical simulations, with good overall agreement. This technique enables the identification and characterisation of objects using their electrical conductivity and magnetic permeability.

Identification of object composition with magnetic inductive tomography

The inductive response of an object to an oscillating magnetic field reveals information about its electrical conductivity and magnetic permeability. Here, we introduce a technique that uses measurements of the angular, frequency, and spatial dependence of the inductive signal to determine the object composition. Identification is performed by referencing an object's inductive response to that of materials with mutually exclusive properties such as copper (high electrical conductivity and negligible magnetic permeability) and ferrite (negligible electrical conductivity and high magnetic permeability). The technique uses a sensor with anisotropic sensitivity to discriminate between the different characters of the eddy current and magnetization driven object responses. Experimental validation of the method is performed using magnetic induction tomography measurement with a radio-frequency atomic magnetometer. Possible applications of the technique in security screening devices are discussed.

Rapid diagnosis and continuous monitoring of intracerebral hemorrhage with magnetic induction tomography based on stacked autoencoder

Magnetic induction tomography (MIT) is a promising approach in rapid diagnosis and continuous monitoring of cerebral hemorrhage. A new algorithm for the reconstruction of intracerebral hemorrhage with MIT, including the location and volume of hemorrhage, is proposed in this study. First, 2D magnetic resonance imaging and computed tomography images of patients with cerebral hemorrhage were used for the development of simulation models. The Stacked Autoencoder (SAE) network was then used to predict the location and volume of hemorrhage by conductivity reconstruction. Finally, the one-dimensional quantitative monitoring index is proposed as an auxiliary diagnostic indicator for assessment of real-time intracranial electrical characteristics. The 2D simulation results showed that the SAE was able to quickly image the location and volume of the hemorrhages. Compared with the back-projection algorithm, the prediction speed of each frame was improved 15-fold, and the accuracy improved by 90.53%. The extracted one-dimensional quantitative monitoring indicators can describe the bleeding status. The diagnostic accuracy and the imaging speed of cerebral hemorrhage were both improved by constructing a realistic head section model and using the proposed SAE network. This research provides a new alternative for dynamic monitoring of hemorrhages and shows the potential advantages of MIT in noninvasive detection.

Research on fuzzy English automatic recognition and human-computer interaction based on machine learning

Fuzzy English recognition is affected by many factors, which leads to certain accuracy problems in intelligent recognition results. In order to improve the automatic recognition efficiency of fuzzy English, based on machine learning technology, this study constructs a neural network model. At the same time, this paper analyzes the research status and existing problems of handwritten character recognition, analyzes the model, and adopts multiple modules for automatic English recognition. In addition, the system is built on the basis of algorithms and model support, which makes fuzzy English recognition intelligent. Finally, in order to study the algorithm and model performance, the fuzzy English recognition is carried out through experiments. The research shows that the model constructed in this paper has certain recognition effect, which can be applied to practice, and can provide theoretical reference for subsequent related research.

Light scattering control in transmission and reflection with neural networks

Scattering often limits the controlled delivery of light in applications such as biomedical imaging, optogenetics, optical trapping, and fiber-optic communication or imaging. Such scattering can be controlled by appropriately shaping the light wavefront entering the material. Here, we develop a machine-learning approach for light control. Using pairs of binary intensity patterns and intensity measurements we train neural networks (NNs) to provide the wavefront corrections necessary to shape the beam after the scatterer. Additionally, we demonstrate that NNs can be used to find a functional relationship between transmitted and reflected speckle patterns. Establishing the validity of this relationship, we focus and scan in transmission through opaque media using reflected light. Our approach shows the versatility of NNs for light shaping, for efficiently and flexibly correcting for scattering, and in particular the feasibility of transmission control based on reflected light.

Sub-picotesla widely tunable atomic magnetometer operating at room-temperature in unshielded environments

We report on a single-channel rubidium radio-frequency atomic magnetometer operating in unshielded environments and near room temperature with a measured sensitivity of 130 fT/ Hz . We demonstrate consistent, narrow-bandwidth operation across the kHz-MHz band, corresponding to three orders of magnitude of the magnetic field amplitude. A compensation coil system controlled by a feedback loop actively and automatically stabilizes the magnetic field around the sensor. We measure a reduction in the 50 Hz noise contribution by an order of magnitude. The small effective sensor volume, 57 mm³, increases the spatial resolution of the measurements. Low temperature operation, without any magnetic shielding, coupled with the broad tunability, and low beam power, dramatically extends the range of potential field applications for our device.

Machine-learning-based atom probe crystallographic analysis

Atom probe tomography is known for its accurate compositional analysis at the nanoscale. However, the patterns created by successive hits on the single particle detector during experiments often contain complementary information about the specimen's crystallography, including structure and orientation. This information remains in most cases unexploited because it is, up to now, retrieved predominantly manually. Here, we propose an approach combining image analysis techniques for feature selection and deep-learning to automatically interpret the patterns. Application of unsupervised machine learning techniques allows to build and train a deep neural network, based on a library generated from theoretically known crystallographic angular relationships. This approach enables direct interpretation of the detector hit maps, as shown here on the example of numerous pure-Al, and is robust enough to function under various conditions of base temperature, pulsing mode and pulse fraction. We benchmark our approach against recent attempts to automate the pattern identification via Hough-transform and discuss the current limitations of our approach. This new automated approach renders crystallographic atom probe tomography analysis more efficient, feature-sensitive, robust, user-independent and reliable. With that, deep-learning algorithms show a great potential to give access to combined atom probe crystallographic and compositional analysis to a large community of users.

Active underwater detection with an array of atomic magnetometers

We report on a 2×2 array of radio-frequency atomic magnetometers in a magnetic induction tomography configuration. Active detection, localization, and real-time tracking of conductive, nonmagnetic targets are demonstrated in air and saline water. Penetration in different media and detection are achieved thanks to the sensitivity and tunability of the sensors, and to the active nature of magnetic induction probing. We obtained a 100% success rate for automatic detection and a 93% success rate for automatic localization in air and water, up to 190 mm away from the sensor plane (100 mm underwater). We anticipate magnetic induction tomography with arrays of atomic magnetometers finding applications in civil engineering and maintenance, the oil and gas industry, geological surveys, marine science, archeology, search and rescue, and security and surveillance.