Modern supervised machine learning for electro-optical and infrared imagery is based on data-driven learning of features and decision making. State-of-the-art algorithms are largely opaque and questions exist regarding their interpretability and generalizability. For example, what are the learned features, what contexts do they work in, and are the algorithms simply memorizing observations and exploiting unwanted correlations or has it learned an internal representation and causal associations that generalize to new environments? Under the hood, current convolutional neural networks (CNN) are sophisticated curve fitters that are sensitive to sampling (volume and variety). This is problematic as collecting data from real systems is often expensive and time consuming. Furthermore, labeling and quality checking of that data can also be prohibitive. As a result, many are looking to augmentation and simulation to efficiently generate more samples. Herein, we focus on ways to combine augmentation and simulation to improve explosive hazard detection. Specifically, we use the Unreal Engine to produce ray traced simulated data sets of environments and emplacements not captured in real data. We also present a new technique, coined altitude modulated augmentation (AMA), that inserts simulated objects into real world background imagery based on metadata to augment new training data. Thus, the goal of AMA is to increase sampling of observed environments. Preliminary results show that the combination of all techniques is best, followed by augmentation, simulation, then real world data.

Numerous real-world applications require the intelligent combining of disparate information streams from sensors to create a more complete and enhanced observation in support of underlying tasks like classification, regression, or decision making. An often overlooked and underappreciated part of fusion is context. Herein, we focus on two contextual fusion challenges, incomplete (limited knowledge) models and metadata. Examples of metadata available to unmanned aerial systems (UAS) include time of day, platform/sensor position, etc., all of which have a potentially drastic impact on sensor measurements and subsequently our decisions derived from them. Additionally, incomplete models limit machine learning, specifically under-sampling of training data. To address these challenges, we investigate contextually adaptive online Choquet integration. First, we cluster and partition the training metadata. Second, a single machine learning model is trained per partition. Third, a Choquet integral is learned for the combination of these models per partition. Fourth, at test/run time we compute the degree of typicality of a new sample to our known contexts. Fifth, our trained integrals are decomposed into a bag of underlying aggregation operators and a new contextually relevant operator is imputed using a combination of the metadata clustering and observation statistics of the integral variables. This process enables machine learning model selection, ensemble fusion, and metadata outlier detection, with subsequent mitigation strategy identification or decision suppression. The above ideas are demonstrated on explosive hazard detection using surrogate data simulated by the Unreal Engine. In particular, the Unreal Engine is used because it provides us with flexibility to explore the proposed ideas across a range of diverse and controlled experiments. Our preliminary results show improved performance for fusion in different contexts and a sensitivity analysis is performed with respect to metadata degradation.

The U.S. Department of Defense developed a 960 pixel laser Doppler vibrometer to capture seismic vibration and detect buried landmines. The sensor relied on a loudspeaker to create the seismic disturbances. This system captured seismic spatial responses while advancing at 1 m/s at 30 m standoff. Automated target recognition algorithms quickly processed the spatial vibration data to warn operators of threats. Sensor noise was highly variable, but on the order of .05mm/s in a band 50–400 Hz over which the sensor operated. Inertial sensing reduced gross Doppler components common to all channels. Even with these corrections, channel dropouts remained a challenge, so filtering of erroneous samples in time and space was required to improve the data quality, which is possible due to 960 simultaneous spatial samples. The large pixel count permitted road scans of 2 m × 1 km in less than 30 min. The system proved effective at finding buried threats that produced seismic anomalies at the surface, but additional challenges lie in differentiating target responses from clutter. This presentation will introduce the system and display high-resolution seismic images excited acoustically. It will also illustrate an acoustic wave captured directly by the lasers as the air's index-of-refraction fluctuated.

A multi-axis heterodyne interferometer concept is under development for observations of five degrees of dynamic freedom using a single illumination source. This paper presents a laboratory system that combines elements of heterodyne Doppler vibrometry, holography, and digital image correlation to simultaneously quantify in-plane translation, out-of-plane rotation, and out-of-plane displacement at the nanometer scale. The sensor concept observes a dynamic object by mixing a single optical field with heterodyne reference beams and collecting these combined fields at the image and Fourier planes, simultaneously. Polarization and frequency multiplexing are applied to separate two segments of a receive Mach-Zehnder interferometer. Different optical configurations are utilized; one segment produces a focused image of the optical field scattered off the object while the other segment produces an optical Fourier transform of the optical field scattered off the object. Utilizing the amplitude and phase from each plane allows quantification of multiple components of transient motion using a single, orthogonal beam.

Detection of landmines at modest distances from a moving vehicle is desired to help protect soldiers from explosive threats. The US Department of Defense developed a sensor system to observe seismic vibration patterns that lead to detection of buried objects in unpaved roads. The system is capable of exciting soil acoustically and observing seismic responses while advancing at 1m/s with a 30m standoff. The optical design and build created 960 simultaneous independent spatial observations that collect data over a 0.5m x 1m area. A gimbal then repositions the sensor to collect a new region. In this sequential manner the system can scan 2m x 1km in less than 30min. Active gimbal stabilization kept beam positions relatively stable on the ground while the vehicle was in motion. Inertial sensing reduced gross Doppler components common to all channels. Even with these corrections, channel dropouts remained a challenge, so filtering of erroneous samples in time and space was required to improve the data quality. Automated target recognition algorithms quickly process the spatial vibration data to warn operators of a threat. We present here an overview of the system and collected data. The system proved effective at finding buried threats that produced seismic anomalies at the surface, but additional challenges lie in differentiating target responses from clutter. The seismic response of naturally occurring environmental clutter in roads produces responses that appear similar to objects of interest, reducing the effectiveness of target detection algorithms.

A multiaxis heterodyne interferometer concept is under development for observations of 5 deg of dynamic freedom using a single illumination source. This Letter presents a laboratory system that combines elements of heterodyne Doppler vibrometry, holography, and digital image correlation to simultaneously quantify in-plane translation, out-of-plane rotation, and out-of-plane displacement. The sensor concept observes a dynamic object by mixing a single optical field with heterodyne reference beams and collecting these combined fields at the image and Fourier planes, simultaneously. Polarization and frequency multiplexing are applied to separate two segments of a receive Mach-Zehnder interferometer. Different optical configurations are utilized; one segment produces a focused image of the optical field scattered off the object while the other segment produces an optical Fourier transform of the optical field scattered off the object. Utilizing the amplitude and phase from each plane allows quantification of multiple components of transient motion using a single, orthogonal beam.

A heterodyne speckle imaging sensor has demonstrated the measurement of three-degrees of vibrational freedom simultaneously. Axial velocity is measured by demodulation, surface velocity gradients are determined by observing speckle translation following extraction from mixed signals.

A laboratory system has demonstrated the measurement of three degrees of vibrational freedom simultaneously through heterodyne speckle imaging. The random interference pattern generated by the illumination of a rough surface with coherent light can be exploited to extract information about the surface motion. The optical speckle pattern is heterodyne mixed with a coherent reference. The recorded optical data is then processed to extract three dimensions of surface motion. Axial velocity is measured by demodulating the received time-varying intensity of high amplitude pixels. Tilt, a gradient of surface displacement, is calculated by measuring speckle translation following extraction of the speckle pattern from the mixed signal. This paper discusses the laboratory sensor concept, signal processing, and experimental results compared with numeric simulations.

Acoustic and seismic waves provide a method to localize compliant mines by vibrating the top plate and a thin soil layer above the mine. This vibration is mostly linear, but also includes a small nonlinear deviation. The main goal of this paper is to introduce a method of processing that uses phase-inversion to observe nonlinear effects in a wide frequency band. The method extracts a nonlinear part of surface velocity from two similar broadcast signals of opposite sign by summing and cancelling the linear components and leaving the nonlinear components. This phase-inversion method is combined with time reversal focusing to provide increased seismic vibration and enhance the nonlinear effect. The experiments used six loudspeakers in a wood box placed over sand in which inert landmines were buried. The nonlinear surface velocity of the sand with a mine compared to the sand without a mine was greater as compared to a linear technique.

Acoustic methods of land mine detection are being developed for nonmetal mines where conventional electromagnetic methods fail. One such method, based on time reversal acoustic principles, is the Phase-inversion method; which detects quadratic and even ordered harmonic nonlinearities in the surface vibration above a mine. While this has shown higher sensitivity than linear methods it cannot detect all nonlinear components. These components can be detected by the Varied Amplitude method that is based on wide band excitation of surface vibration using the principles of time reversal acoustics. The nonlinear eects are measured for surface vibrations with dierent amplitudes and the normalized received signals can then be subtracted to reveal all the nonlinear components.. We conducted a set of experiments using a box with six loudspeaker and an array of spatially distributed geophones to measure the linear and nonlinear components for various types of mines in dierent soils. Information about nonlinear components’ spatial distribution allows one to observe greater mine/no mine contrast than with linear and other nonlinear techniques. [Work was supported by the U.S. Army RDECOM CERDEC Night Vision and Electronic Sensors Directorate.]

Acoustic sniper localization algorithms are well established for open field detection; however, these algorithms are less accurate in urban environments due to reverberation and diffraction. Research is being performed to understand the effects of obstructions on shock wave propagation. Barrier attenuation calculations that use Fresnel number to characterize obscurity will be compared to measured attenuation of shock wave frequency components. An artificial building was set up and shock waves from bullets were recorded by microphones with varying degrees of obscurity. The experimental setup allowed for a range of Fresnel numbers between 5 and 11 resulting in attenuation up to 30 dB. Attenuation increased as obscurity and Fresnel number increased up to a threshold. Beyond this point the magnitude of attenuation decreases in both theory and experiment. These data support the relationship between attenuation and microphone obscurity and may provide a means to estimate the amplitude and N-wave slope of the shock wave prior to obstruction.

Time reversal focuses seismic waves and excites nonlinear surface vibrations that are large in the presence of a landmine. These soil vibrations provide an accurate localization cue when contrasted with measurements made without a mine. Traditionally nonlinear effects were investigated by measuring the interaction of harmonic waves and nonlinear resonance frequency shifts. These methods provide high signal to noise ratios, but required tuning of excitation frequencies resulting in test times unsuitable for field applications. Time reversal is capable of providing broad band excitation at amplitudes large enough to drive the nonlinear mechanisms at measurable levels while reducing the overall measurement time. Spatial and temporal focusing data will be presented for three different time reversal systems compromising speakers and vibrators for seismic excitation and a Doppler laser vibrometer and geophones for signal recording. The high amplitude signals generated by these systems adequately drive the nonlinear responses. Special processing techniques, phase inversion and varied amplitude, have been developed to extract the nonlinear responses from the broadband response by cancelling the linear part. The remaining nonlinear response shows greater mine/no mine contrast than the linear data. Practical implementations of the developed methods will be discussed.

Acoustic methods of land mine detection are being developed for nonmetal mines where conventional electromagnetic methods fail. One such method, based on time reversal acoustic principles, is the hhase-inversion method; which detects quadratic and even ordered harmonic nonlinearities in the surface vibration above a mine. While this has shown higher sensitivity than linear methods it cannot detect all nonlinear components. These components can be detected by the Varied Amplitude method that is based on wide band excitation of surface vibration using the principles of time reversal acoustics. The nonlinear effects are measured for surface vibrations with different amplitudes and the normalized received signals can then be subtracted to reveal all the nonlinear components. We conducted a set of experiments using a box with six loudspeaker and an array of spatially distributed geophones to measure the linear and nonlinear components for various types of mines in different soils. Information about nonlinear components' spatial distribution allows one to observe greater mine/no mine contrast than with linear and other nonlinear techniques. [Work was supported by the U.S. Army RDECOM CERDEC Night Vision and Electronic Sensors Directorate.]

Landmine detection is tactically important for main supply route clearance in counter-insurgency operations. Seismo-acoustic detection is one of multiple technologies being pursued by the Army in an effort to address this need. The experiments to be presented were designed to address the use of phase data for landmine detection. These experiments are, in large part, a replication of the efforts by T Wang et al. [Proc. SPIE 5415, 70–79 (2004)]. In the present work, phase and magnitude data are used to detect three targets: a V-S 1.6 anti-vehicular mine (easy target), an M-19 anti-tank mine (difficult target), and a resonant plate (very easy control). These were tested in four situations: free space, flush bury, 1 bury, and 2 bury. The soil surface velocity was measured using an array of four laser Doppler vibrometers, and the acoustic pressure used for excitation was recorded with a microphone. In certain instances, the relative phase between the surface vibration and acoustic pressure directly above the mine was different from soil alone.

Acoustic sniper localization algorithms have proven useful to soldiers, but are difficult to implement in urban environments, due to reverberation and diffraction. Research is being performed to assist traditional localization methods by interpreting reverberant responses present in urban environments. Measured sniper responses are processed to establish if the first acoustic path is direct or indirect, based on characteristics of the received signal. If the first arrivals on several receivers are direct, an accurate relative time delay can be determined for each, which leads to accurate source localization. Indirect microphone arrivals can adversely affect the localization algorithm. Various metrics were used to characterize received signals: Arrival energy, reverberation decay time, initial arrival crest factor, and relationships between first and subsequent arrivals. Combining these factors aids in the identification of obstructed receivers, providing information about the effectiveness of a channel in a localization algorithm.

The nonlinear vibration response of land mines is an effective detection clue. This response can be measured by application of two frequency acoustic waves producing a vibration spectrum above the mine that is different from its surrounding. As an alternative to this approach, this work investigates time reversal techniques as a method of acoustically exciting nonlinear mine vibrations over a broad frequency range. The phase-inversion method used here requires two short broadband time-reversed focused signals with opposite sign, i.e. phase inverted. The focused signal and the inverted focused signal are broadcast sequentially and the responses are summed in post processing. This cancels the linear response leaving the nonlinear component. A theoretical model has been developed that treats the mine stiffness as a nonlinear spring. It predicts that the dominant nonlinear frequencies correspond to the second harmonics of the mine’s primary resonance. Experiments were conducted using six loudspeakers in a box placed directly over a mine. The measured responses confirm most of the model predictions, and illustrate differences between the soil and mine nonlinearities. Work was supported by the U.S. Army RDECOM CERDEC Night Vision and Electronic Sensors Directorate

Lasers provide one method of remotely exciting propagating waves in soil. The absorption of optical energy results in plasma formation, a dynamic change near the soil surface, and excitation of propagating waves. Experiments were performed using a high power CO2 pulsed laser to determine the pressure waveforms generated in desert sand and caliche clay. The laser energy ranged from 400–750 J with a pulse duration of 35 μs. The resulting pressure pulse was observed for duration of at least 400 μs and had a peak pressure of 100 kPa. The duration increased at greater depths, while the peak pressure and energy in a pulse decreased rapidly at these depths. Surface velocity reached a peak of 35 mm/s as acquired on geophones placed in proximity to the laser plasma event. This system, in combination with a laser Doppler vibrometer, has the potential to be a useful component of non-invasive mine detection equipment.

Acoustic methods of land mine detection are being developed for nonmetal mines where conventional electromagnetic methods fail. Nonlinear acoustic methods have shown higher sensitivity than linear methods. In previous studies, the nonlinear methods, based on interaction of two frequency waves and phase inversion of two pulses, were investigated. We propose an alternative nonlinear method based on the use of wide-frequency pulses with different amplitude and subtraction of the normalized received signals. This method evaluates the nonlinear effects in a wide-frequency band similar to the phase inversion method and it detects nonlinear components that are lost in the phase inversion method. In our tests, the suggested method was combined with time reversal acoustic focusing to concentrate seismic wave energy in proximity to the mine, which increased nonlinear effects. Experiments with various types of mines were conducted at Fort Belvoir using six loudspeakers in a box for excitation in the frequency band 50–500 Hz. The nonlinear mine response measured by the varied amplitude method provides the highest mine/no mine contrast among all the experiments performed using linear and nonlinear techniques. [Work was supported by the U.S. Army RDECOM CERDEC Night Vision and Electronic Sensors Directorate.]

Localization of an acoustic blast source in urban battle spaces is frequently complicated by multiple arrivals resulting from reflections within the acoustic environment. Temporal windowing can be used to isolate the direct arrival in situations where a direct path between source and receiver exists. However, if the source is located around a corner from a receiver, the first arrival would lead to an erroneous source location estimate. These complications have prompted the authors to perform initial measurements of reverberation in a semi-urban environment. These measurements will assess the salience of reverberation as a means for estimating range to a blast source.

A land mine detection system has been developed based on excitation of seismic waves by four mechanical shakers and measurements of ground vibration by a 16-beam, scanning laser Doppler vibrometer (LDV). Time-reversal acoustic (TRA) focusing was applied to concentrate seismic waves using the above system. Linear swept frequency signals and orthogonal initial signals in the frequency band of 100–500 Hz were used in outdoor testing. The system impulse response from any shaker to the LDV output was computed by cross correlating the original and recorded signals. With orthogonal signals, the system impulse responses between multiple shakers and the LDV were measured simultaneously, thus reducing the overall measurement time. Each shaker then reradiates the time-reversed impulse responses focusing the seismic energy to a point on the ground. Experiments demonstrate that the TRA focusing provided a high concentration of elastic wave energy in the test area with typical focusing spots about 20 cm in diameter. [Work supported by U.S. Army RDECOM CERDEC Night Vision and Electronic Sensors Directorate.]