The instrument, which the researchers call a computer vision-based measurement monitoring (CVMM) system, works on a principle reminiscent of a surveyor’s theodolite. A 20-centimeter chessboard target is fixed to the point of interest on the dam crest, while the camera sits at a stable vantage point on the opposite riverbank. OpenCV algorithms running in Python detect the corners of the chessboard in each frame, calibrate for lens distortion, and convert pixel shifts into real-world displacement using the mathematics of planar homography and camera projection matrices. Because the true size of the chessboard is known, the system can even estimate the distance between the camera and the target from how large the board appears in the image, recovering motion along the depth axis as well as horizontally and vertically.

Accuracy was established through careful laboratory characterization. On a three-axis manual stage, the static standard deviation of the measurements was just 0.058, 0.042, and 0.067 pixels in the X, Y, and Z axes respectively, across more than 2,500 data points per axis. When the chessboard was moved by exactly one centimeter, the system reproduced the motion with an accuracy of 0.01 centimeters, where a single pixel corresponded to roughly 0.16 centimeters at the test distance. Long-range trials at 15, 20, 35, and 50 meters showed average standard deviations of 0.01 to 0.02 centimeters up to 35 meters, degrading to 0.06 centimeters beyond 50 meters, which led the team to recommend installation distances under 35 meters for their industrial lens with 10x optical zoom.

Crucially, the system does more than track displacement. Running at 100 frames per second—nearly the maximum 120 frames per second supported by the camera module—the CVMM captures vibrational motion that conventional time-lapse photography would average away. Displacement time series can be differentiated to yield velocity, and then decomposed spectrally. The researchers applied the Fast Fourier Transform to isolate dominant frequencies, the Short-Time Fourier Transform to see how those frequencies evolve in time, and finally the Hilbert-Huang Transform, a technique designed for nonlinear, non-stationary signals that breaks the record down into intrinsic mode functions and computes instantaneous frequency and energy for each one.

To validate the vibration measurements, the team mounted a chessboard alongside a commercial Raspberry Shake 4D seismograph on a QUANSER shake table and drove the table through a series of controlled oscillations at constant 1-centimeter amplitude but varying accelerations, from 0.8 to 200 gal. The image acquisition frame rate achieved 98.76 percent of its 100-frames-per-second target, and image recognition succeeded 99.70 percent of the time. After a median filter smoothed out spike noise, spectral analysis showed that the primary frequencies detected by the CVMM aligned closely both with the shake table’s input settings and with the independent seismograph record. The maximum amplitude error was 0.13 centimeters, with an average of 0.09 centimeters, confirming that a camera tracking a checkerboard can genuinely function as a non-contact vibrometer across a range from 0.1 to 50 Hz.

The decisive test came on the Lantao river in Nantou County, Taiwan, where the researchers built an artificial earth dam 27.8 meters wide and 3.0 meters high from a mix of boulders, gravel, and sand typical of landslide-formed debris dams. The chessboard was planted on the dam crest near a pre-cut breaching channel, and the CVMM instrument was set up 35.32 meters away on the right riverbank, a distance verified with a theodolite. Water was then fed into the reservoir until it overtopped the crest and carved the dam apart. Over 1,249 seconds of monitoring, the target slid 14.64 centimeters horizontally toward the developing breach and settled 26.77 centimeters vertically before it finally collapsed with the failing dam.

The time-frequency analysis of this field data revealed a striking pattern. As the breach developed between 15:16 and 15:20, vibration energy surged across three distinct frequency bands: a high-frequency band from 20 to 50 Hz, a medium band from 10 to 20 Hz, and a low band below 10 Hz. Following established findings that high-frequency ground vibrations arise from small particles like sand and gravel colliding while low frequencies reflect the movement of larger rocks, the team concluded that the dominant energy came from the washing out and collision of finer sediments as overflow eroded the crest. The vertical axis carried more energy than the horizontal—maximum instantaneous energies of 2,500, 350, and 50 square centimeters per second squared in the high, medium, and low bands of the Y axis, compared with 1,800, 200, and 30 on the X axis—consistent with settlement being the more violent process.

Most importantly for disaster response, the Hilbert spectra showed a sharp rise in instantaneous energy beginning around 15:16, roughly two minutes before the overflow at 15:18 and the full breach at 15:20. Because the first intrinsic mode function carried over 84 percent of the total energy, the researchers proposed an early-warning threshold of 1,000 square centimeters per second squared in that high-frequency band, a level that was exceeded more than 36 times during the initial breach phase. Combined with a simpler displacement indicator—settlement exceeding 5 centimeters at the moment of overtopping—these thresholds could give downstream communities precious minutes of advance notice. LiDAR scans taken before and after the test independently confirmed the elevation changes the camera system had tracked in real time.

The economics are equally compelling. The entire CVMM instrument costs roughly 1,000 to 2,000 US dollars depending on configuration, about one-fifth the price of conventional monitoring packages such as total stations, GNSS receivers, or LiDAR units. If the chessboard target is destroyed in a failure event, replacing it is trivial. The system also delivers three data streams simultaneously—imagery, displacement, and vibration—where each traditional device provides only one, and it requires no physical contact with a structure that may be on the verge of collapse, addressing a fundamental weakness of contact sensors like the Raspberry Shake, which cannot safely be deployed on a failing slope.

The authors are candid about the limitations. The energy threshold was derived from a single field test on one soil type and will need validation across different dam materials. Computer vision remains vulnerable to heavy rain, poor night lighting, atmospheric turbulence, and camera shake, although previous work by the same group using a weatherproof enclosure and a solar-powered LED chessboard achieved 24-hour monitoring with errors near 0.1 centimeter at 50 meters. No Raspberry Shake was deployed on the dam crest itself for direct comparison during the breach, and coupling the vibrational signatures to pore pressure changes and internal failure mechanisms will require additional instruments such as piezometers. Even so, the demonstration stands as a proof of concept that a modest camera, a printed checkerboard, and clever signal processing can transform dam safety monitoring—turning an ordinary machine vision pipeline into an early-warning sentinel for one of the deadliest failure modes in mountainous terrain.

Subject of Research: Computer vision-based monitoring of dam breach displacement and vibration for early warning

Article Title: Computer vision-based system of vibration monitoring for early-warning signals in dam breach

Article References: Chen, I.-H., Chen, S.-C., & Yang, R.-J. (2026). Computer vision-based system of vibration monitoring for early-warning signals in dam breach. Environmental Earth Sciences, 85(15), Article 392. https://doi.org/10.1007/s12665-026-13117-7

Image Credits: AI Generated

DOI: 10.1007/s12665-026-13117-7

Keywords: computer vision, dam breach, vibration monitoring, displacement monitoring, Hilbert-Huang Transform, OpenCV, Raspberry Pi, early warning, time-frequency analysis, earth dam, landslide dam, structural monitoring

The core challenge the method confronts is known in the field as sparse reconstruction. In a typical monitoring campaign, accelerometers or strain gauges are placed at a limited number of accessible locations, often far fewer than the number needed to describe the structure’s motion completely. Between the measured points lies a vast space of unobserved behavior, and traditional interpolation or modal expansion techniques struggle to fill it accurately, particularly when the structure is excited by unpredictable forces such as wind, traffic, or seismic shaking. Classical approaches usually assume that the response can be described by a fixed set of vibration modes with slowly varying amplitudes, an assumption that breaks down under nonlinear behavior, transient impacts, or rapidly changing operating conditions.

The innovation at the heart of the new work lies in the way the model processes signals. Rather than treating a vibration record as a simple sequence of numbers sampled in time, the time-frequency gated transformer first transforms the data into a representation that captures both temporal and spectral information simultaneously. This matters because structural responses are inherently multiscale: a bridge deck may carry slow, low-frequency sways driven by wind alongside rapid, high-frequency ripples generated by passing vehicles or local impacts. A model that looks only at the raw time series can miss the spectral signatures that distinguish one excitation mechanism from another, while a purely frequency-domain view loses the precise timing of transient events. By operating in the joint time-frequency domain, the network gains access to a richer description of the physics encoded in the measurements.

The transformer architecture, which has transformed natural language processing and computer vision over the past several years, provides the computational backbone. Transformers rely on attention mechanisms, mathematical operations that allow the model to weigh the relevance of every part of an input sequence when interpreting any given part. Applied to structural data, attention lets the network learn long-range spatial and temporal correlations: for example, how a vibration pattern measured at the base of a tower relates to the response at its top, or how an impact at one point in time influences the oscillations observed seconds later. This ability to capture dependencies across long distances in the data is precisely what sparse reconstruction demands, since unmeasured locations must be inferred from patterns observed elsewhere on the structure.

What distinguishes this implementation from a vanilla transformer is the gating mechanism applied in the time-frequency domain. Gates are learned functions that selectively amplify or suppress components of the representation, allowing the network to decide, on a case-by-case basis, which frequency bands and time windows carry the most reliable information for the reconstruction task at hand. During training, the model adjusts these gates so that noise-dominated or physically irrelevant portions of the signal are downweighted, while informative features pass through to the reconstruction layers. The result is a form of learned signal filtering that adapts dynamically to the data rather than relying on hand-designed filters, a significant advantage when dealing with real-world measurements contaminated by sensor noise, environmental variability, and electromagnetic interference.

The practical implications of such a system are considerable. Structural health monitoring has become a global priority as aging bridges, dams, and buildings face increasing loads from climate extremes, heavier traffic, and material degradation. Full-field reconstruction, the ability to estimate the displacement, velocity, or acceleration at every point of a structure from limited measurements, enables engineers to detect damage earlier, locate it more precisely, and assess remaining service life with greater confidence. In aerospace, where instrumenting every square meter of a wing or fuselage is impractical, a model that infers the complete dynamic response from a sparse sensor array could improve flutter prediction, fatigue tracking, and certification testing. In mechanical engineering, rotating machinery such as wind turbines and jet engines could be monitored with fewer sensors, reducing cost and downtime.

Deep learning approaches to this problem have been attempted before, but they have often stumbled on generalization. A network trained on one structure, one sensor layout, or one class of excitation frequently fails when conditions change, which is a serious limitation for infrastructure that must endure decades of varying environments. The time-frequency gating strategy is aimed squarely at this weakness. By explicitly separating information across time and frequency scales and learning which combinations matter, the model builds a more physically grounded internal representation, one that is less tied to the specific idiosyncrasies of its training data. The authors report that the architecture reconstructs dynamic responses with high accuracy across a range of scenarios, outperforming conventional baseline methods that lack the joint time-frequency treatment.

Like any data-driven method, the approach depends on the quality and diversity of its training data, and the researchers acknowledge that deploying such models on real structures requires careful validation against measured ground truth. Questions of uncertainty quantification also remain active areas of research: engineers need to know not only what the model predicts but how confident it is in each estimate, particularly when the reconstruction informs safety-critical decisions. The field is moving rapidly toward hybrid frameworks that combine the flexibility of neural networks with the guarantees of physics-based models, and gated time-frequency transformers of the kind presented here are likely to serve as powerful components within such systems. The study’s publication in a Nature Portfolio engineering journal signals growing mainstream recognition that machine learning and structural dynamics have converged into a productive research frontier.

The broader significance of the work extends beyond any single application. It exemplifies a wider trend in which attention-based architectures, originally developed for language, are being repurposed to solve problems in physical science and engineering, from weather forecasting to materials discovery. Structural dynamics, with its rich multiscale character and its sparse, noisy measurements, turns out to be a natural fit for these tools. As sensor hardware becomes cheaper and wireless networks make dense instrumentation feasible, the bottleneck will shift from data acquisition to data interpretation, and models like the time-frequency gated transformer will define how effectively that bottleneck is cleared. For the engineers responsible for keeping the world’s infrastructure safe, the study offers a glimpse of monitoring systems that see far more than the sensors they are built on, reconstructing the hidden motion of structures with a fidelity that was recently the province of simulation alone.

Subject of Research: Deep learning-based reconstruction of structural dynamic responses from sparse vibration measurements

Article Title: Time-frequency gated transformer for structural dynamic response reconstruction

Article References: Song, X., Yang, F., Li, R., Ma, X., Wang, Y., Wang, S., Deng, Q., & Zheng, S. (2026). Time-frequency gated transformer for structural dynamic response reconstruction. Communications Engineering. https://doi.org/10.1038/s44172-026-00788-0

Image Credits: AI Generated

DOI: 10.1038/s44172-026-00788-0

Keywords: structural health monitoring, transformer, deep learning, time-frequency analysis, vibration reconstruction, sparse sensors, structural dynamics, attention mechanism, civil infrastructure, aerospace engineering, machine learning, signal processing

Conventional approaches to unknown waveform detection have generally fallen into two camps. Statistical anomaly detection methods build a model of what normal signals look like and flag anything that deviates from it, an approach that is conceptually elegant but notoriously brittle in cluttered spectral environments where legitimate signals vary wildly in power, modulation and bandwidth. Deep learning classifiers, meanwhile, have delivered impressive results on benchmark datasets of known modulations, yet they stumble when confronted with waveform families absent from their training sets. Some researchers have tried to bridge this gap with generative techniques that synthesize artificial examples of unknown signals, but generating realistic synthetic samples is itself an unsolved problem, and poorly generated data can bias a model in ways that are difficult to diagnose. The result, in practice, is that fielded systems often fail exactly when they are needed most: during encounters with genuinely novel emitters.

The new discriminative model sidesteps synthetic sample generation altogether. Instead of trying to imagine what unknown waveforms might look like, it learns a richer description of the signals it does know, so that anything sufficiently different stands out sharply. The key architectural decision is the joint use of time-domain and frequency-domain representations. The time domain captures how a signal’s amplitude evolves moment to moment, preserving transient features, timing structure and modulation transitions that unfold in sequence. The frequency domain, obtained through transforms such as the Fourier transform, reveals how energy is distributed across the spectrum, exposing carrier offsets, spectral occupancy, harmonic structure and bandwidth fingerprints. Human signal analysts have long toggled between these two views on oscilloscopes and spectrum analyzers; the invention encodes that dual perspective directly into the learning pipeline.

But merely concatenating two views of a signal would not, by itself, guarantee better detection. The second pillar of the invention is a cosine similarity loss function that reshapes the model’s internal feature space. In machine learning, a loss function defines what a model is penalized for getting wrong, and therefore what it learns to prioritize. Cosine similarity measures the angle between two feature vectors rather than the distance between them, meaning it is sensitive to the direction of a representation but insensitive to its magnitude. By training with a cosine similarity objective, the system is pushed to align the feature vectors of signals from the same class more tightly while steering vectors of different classes apart. Class-specific features become more cleanly separated, and the model develops a sharper decision boundary between familiar waveform families and everything else.

The practical consequence of this design is a measurable jump in performance. In testing against comparable models that lacked the combined time-frequency representation and the cosine similarity mechanism, the invention delivered approximately a 10 percent improvement in detection accuracy for unknown waveforms. A ten percent gain may sound incremental, but in the context of unknown-signal detection, where baseline systems frequently operate in regimes of unreliable performance, it represents a substantial margin. It means fewer missed detections of genuinely anomalous emitters and fewer false alarms triggered by ordinary variations in known signals, both of which carry real operational costs. The improvement stems directly from the model’s enhanced ability to differentiate subtle waveform variations that do not appear in its training data, rather than from any increase in raw computational capacity.

Robustness and generalization are the qualities that make this improvement durable rather than dataset-specific. Because the model’s features are aligned by direction in the feature space, they are less sensitive to the scale of a signal’s power, a property that matters enormously in realistic radio environments where the same emitter may be received at wildly different strengths depending on distance, terrain and antenna orientation. The dual-domain representation also provides redundancy: a waveform feature that is ambiguous in the time domain, such as a slight shift in spectral occupancy, may be unmistakable in the frequency domain, and vice versa for temporal phenomena. This built-in cross-checking gives the system a form of resilience that single-representation classifiers lack, allowing it to maintain accurate classification under the noisy, adversarial conditions that characterize contested spectrum.

The anticipated applications span both military and civilian domains. In electronic warfare, the ability to detect and classify unknown communication signals is foundational to situational awareness, since an adversary’s new emitter is by definition absent from any pre-existing threat library. Spectrum management authorities could deploy the technology to monitor and enforce the use of the electromagnetic spectrum, identifying rogue transmissions and interference sources that conventional monitoring tools miss. Intelligence, surveillance and reconnaissance operations depend on reliable waveform identification, and the invention’s resistance to the biases introduced by synthetic sample generation makes it a more trustworthy analytic tool. Radio astronomers, who fight a constant battle against radio-frequency interference contaminating observations of faint cosmic sources, could use the model to detect and mitigate intruding signals. Communication security systems, meanwhile, could apply it to flag unauthorized or anomalous transmitters operating within protected networks.

The technology is at technology readiness level 3, the stage at which a concept has been proven analytically and experimentally in laboratory conditions but has not yet been integrated into an operational prototype. It is patent pending and available for licensing through the Research Foundation for the State University of New York, the nation’s largest research foundation, which supports research across the SUNY system in areas including artificial intelligence for the public good, quantum technologies, next-generation semiconductors, biotech and medicine, and energy and climate solutions. SUNY, the largest comprehensive system of higher education in the United States, oversees nearly a quarter of academic research in New York, with research expenditures of nearly 1.5 billion dollars in fiscal year 2025, and the foundation offers multiple pathways for translating such innovations into commercial and economic development opportunities.

What makes the invention notable beyond its immediate applications is the clarity of its underlying insight. Much of modern machine learning progress has come from adding scale: more layers, more parameters, more data. This work demonstrates that careful attention to how a signal is represented, and to how a model is trained to organize its internal features, can yield decisive gains without any of that overhead. By combining two complementary mathematical descriptions of the same physical phenomenon and aligning them with a geometry-aware loss function, the researchers have built a system that sees the electromagnetic world the way an experienced analyst does, from multiple angles at once. As the radio spectrum grows more crowded and the population of uncooperative or unrecognized transmitters continues to expand, tools that can reliably distinguish the known from the genuinely unknown will only grow in importance, and this time-frequency approach offers a concrete, testable step toward that goal.

Subject of Research: An AI model combining time-domain and frequency-domain features with cosine similarity loss for detecting unknown electromagnetic waveforms.

Article Title: Contrasting time-frequency representations for unknown waveform detection

Article References: Contrasting time-frequency representations for unknown waveform detection. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: unknown waveform detection, cosine similarity loss, time-frequency analysis, electromagnetic spectrum, machine learning, electronic warfare, spectrum management, radio-frequency interference, signal classification, deep learning, radio astronomy, SUNY licensing

Researchers led by Balu P. Bhusari of the Ramrao Adik Institute of Technology, DY Patil Deemed to be University in Navi Mumbai, together with colleagues including Akshay A. Jadhav of SIES Graduate School of Technology, have developed a Dual-Path Attention-Augmented Residual Network, abbreviated DA-ResNet. The model is built on a ResNet-50 backbone, a widely used convolutional neural network architecture known for its ability to train very deep layers without suffering from vanishing gradients. What sets the new approach apart is not raw depth but a series of carefully engineered modules that address the specific physics of low Jamming-to-Noise Ratio, or JNR, conditions, where the distinguishing features of a jamming waveform are largely obscured by noise.

The problem the team set out to solve is fundamentally one of feature extraction. Radar jamming signals come in many varieties, including deceptive jamming that mimics real target returns, barrage jamming that blankets entire frequency bands, and compound jamming that combines multiple techniques. Each type leaves a characteristic fingerprint in the time-frequency domain, the two-dimensional representation of how a signal’s energy is distributed across frequency over time. When the jamming-to-noise ratio is high, these fingerprints are crisp and easy for machine learning models to read. As JNR drops toward and below zero decibels, however, the fingerprints fade into the noise floor, and models trained on cleaner data begin to fail dramatically.

To combat this, the researchers equipped their network with a Gabor-initialized convolutional front-end. Gabor filters are mathematical functions long prized in signal and image processing for their ability to capture oriented textures and periodic patterns, precisely the structures that jamming waveforms leave in spectrogram images. By initializing the first convolutional layer with Gabor filters rather than random weights, the network starts with an innate sensitivity to the directional ridges and oscillatory patterns that distinguish one jamming type from another, even when those patterns are faint. This biologically inspired initialization, which echoes the receptive fields found in the human visual cortex, gives the model a head start that random initialization cannot provide.

The second key ingredient is attention. The architecture incorporates Squeeze-and-Excitation, or SE, attention blocks, which perform channel-wise feature recalibration. In practical terms, the network learns to weigh the importance of each feature channel dynamically, amplifying the channels that carry discriminative information about the jamming type and suppressing those dominated by noise. This adaptive suppression is critical in low JNR regimes, where a substantial fraction of the raw input energy is useless or misleading. Rather than treating every feature equally, the model effectively learns which parts of the signal representation deserve scrutiny and which should be tuned out.

The third innovation is a hybrid dual-path pooling strategy at the end of the network. Most convolutional classifiers collapse their final feature maps using global average pooling, which summarizes the overall energy distribution of a signal but can wash out brief, sharp events. Global max pooling does the opposite, preserving the strongest transient peaks but ignoring the broader energy structure. The DA-ResNet combines both, running the features through parallel paths that capture global energy distribution and transient peak characteristics simultaneously. Because different jamming types differ in different ways, some through their overall spectral shape and others through sudden spikes or chirps, this dual-path design ensures that neither kind of evidence is lost before classification.

The team evaluated the model on two independent datasets containing time-frequency representations of radar jamming signals, with jamming-to-noise ratios spanning an unusually demanding range from minus 10 decibels to plus 30 decibels. The results were striking. DA-ResNet achieved an overall recognition accuracy of 97.07 percent, outperforming existing state-of-the-art methods across the tested conditions. Importantly, the model demonstrated strong cross-SNR generalization, meaning it retained high accuracy when tested on signal conditions different from those seen during training, a property that matters enormously in real deployments where the interference environment cannot be predicted in advance. The authors also note that the model achieves this performance with moderate computational overhead, an important consideration for radar systems where processing latency and hardware budgets are constrained.

Beyond raw accuracy, the researchers subjected their network to a battery of diagnostic analyses designed to confirm that it was learning meaningful, separable representations rather than exploiting artifacts. Confusion matrices showed clean separation among jamming classes with limited cross-confusion. Receiver operating characteristic curves quantified the trade-off between detection sensitivity and false alarms across classes. A t-SNE visualization, a technique that projects high-dimensional feature embeddings into two dimensions, revealed tightly clustered groups corresponding to individual jamming types, indicating that the learned features are genuinely discriminative. Perhaps most compelling for practitioners, Grad-CAM heatmaps highlighted the specific regions of the time-frequency images that the network attended to when making its decisions, offering a form of explainable artificial intelligence that builds trust in the model’s judgments and helps engineers verify that the network is focusing on physically meaningful signal structures.

The significance of this work extends beyond a single benchmark. Electronic warfare is escalating as a domain of geopolitical competition, and the ability of a radar to recognize the jamming it faces in real time determines which countermeasures it can deploy. A radar that can correctly identify a deceptive sweep jamming at minus 10 decibels JNR can switch to appropriate anti-jamming processing, such as sidelobe blanking, frequency agility, or adaptive beamforming, before the deception succeeds. Models like DA-ResNet, which remain robust deep into the noise floor, could therefore translate directly into survivability advantages for both military platforms and critical civilian infrastructure such as air traffic control and GNSS-dependent navigation, where jamming and interference are growing concerns.

The study also reflects broader trends in applied machine learning. Rather than inventing an entirely new network family, the authors combined proven components, residual learning, Gabor-based initialization, squeeze-and-excitation attention, and dual pooling, into an architecture tailored to the physics of the problem. This pattern of physics-informed deep learning, where domain knowledge shapes the inductive biases of the model, is proving especially valuable in signal processing fields where training data is limited and noise is adversarial. The researchers, who received no external funding for the work and report no competing interests, suggest that their framework’s balance of accuracy, robustness, and computational efficiency makes it a practical candidate for integration into next-generation electronic counter-countermeasure systems, where every decibel of recognition capability recovered from the noise could make the difference between a radar that sees through the storm and one that is blinded by it.

Subject of Research: Deep learning-based recognition of radar jamming signals under low jamming-to-noise ratio conditions

Article Title: Dual-Path Attention-Augmented ResNet for Robust Radar Jamming Recognition under Low JNR Conditions

Article References: Bhusari, B. P., Jadhav, A. A., Somani, S., More, S., & Patil, S. (2026). Dual-Path Attention-Augmented ResNet for Robust Radar Jamming Recognition under Low JNR Conditions. Mobile Networks and Applications. https://doi.org/10.1007/s11036-026-02550-4

Image Credits: AI Generated

DOI: 10.1007/s11036-026-02550-4

Keywords: radar jamming recognition, deep learning, ResNet-50, attention mechanism, Squeeze-and-Excitation, Gabor filters, time-frequency analysis, low JNR robustness, electronic counter-countermeasures, explainable AI, Grad-CAM, dual-path pooling