Event Classification by Physics-informed Inpainting for Distributed Multichannel Acoustic Sensor with Partially Degraded Channels

TL;DR

The paper tackles robust sound event classification in distributed multichannel acoustic sensing when channels are degraded and sensor layouts vary between training and testing. It introduces a learning-free, physics-informed frontend based on reverse time migration to form a scene image from multichannel spectrograms, followed by forward projection to inpainted signals, log-mel feature extraction, and Transformer-based classification. The approach yields robust performance across circular, linear, and right-angle layouts, notably improving accuracy on the right-angle layout by 13.1 percentage points, and shows that channel weights align with SNR rather than raw geometry, indicating effective leveraging of channel quality. This method complements learning-based SEC by incorporating physical priors, enabling layout-open DMAS with severe channel degradation and offering a practical path toward scalable environmental sound sensing.

Abstract

Distributed multichannel acoustic sensing (DMAS) enables large-scale sound event classification (SEC), but performance drops when many channels are degraded and when sensor layouts at test time differ from training layouts. We propose a learning-free, physics-informed inpainting frontend based on reverse time migration (RTM). In this approach, observed multichannel spectrograms are first back-propagated on a 3D grid using an analytic Green's function to form a scene-consistent image, and then forward-projected to reconstruct inpainted signals before log-mel feature extraction and Transformer-based classification. We evaluate the method on ESC-50 with 50 sensors and three layouts (circular, linear, right-angle), where per-channel SNRs are sampled from -30 to 0 dB. Compared with an AST baseline, scaling-sparsemax channel selection, and channel-swap augmentation, the proposed RTM frontend achieves the best or competitive accuracy across all layouts, improving accuracy by 13.1 points on the right-angle layout (from 9.7% to 22.8%). Correlation analyses show that spatial weights align more strongly with SNR than with channel--source distance, and that higher SNR--weight correlation corresponds to higher SEC accuracy. These results demonstrate that a reconstruct-then-project, physics-based preprocessing effectively complements learning-only methods for DMAS under layout-open configurations and severe channel degradation.

Figures (6)

• Figure 1: Concept of multichannel signals with partially degraded channels and RTM inpainting
• Figure 2: Example of RTM-inpaint spectrogram
• Figure 3: (a) Coordinate and (b) distance distribution
• Figure 4: Accuracy [%] of SEC for each layout
• Figure 5: Visualization of spatial weight (Eq. \ref{['eq:spatial_weight']}) for each method