# Material Classification from Non-Line-of-Sight Acoustic Echoes Using Wavelet-Acoustic Hybrid Feature Fusion

## TL;DR

A new model using wavelet and acoustic features can accurately classify materials from sound echoes even when direct sound paths are blocked, with potential applications in robotics and monitoring.

## Contribution

A novel wavelet–acoustic hybrid feature fusion method with CNN–LSTM architecture achieves 99% accuracy in non-line-of-sight material classification.

## Key findings

- The Wavelet–Hybrid CNN–LSTM model achieved 99% balanced accuracy and macro-F1 score in material classification.
- SHAP analysis showed MFCC and wavelet entropy-energy features complement each other in material discrimination.
- The model enables interpretable and real-time acoustic sensing in non-line-of-sight environments.

## Abstract

What are the main findings?
The proposed Wavelet–Hybrid CNN–LSTM model achieved 99% balanced accuracy and macro-F1 score in classifying materials from non-line-of-sight (NLOS) acoustic echoes, outperforming wavelet-only and classical acoustic models.SHAP-based explainability analysis revealed that Mel-Frequency Cepstral Coefficient (MFCC) and wavelet entropy-energy features play complementary roles in material discrimination, allowing the model to not only classify with high accuracy but also interpret physical material properties such as hardness, density, and porosity.

The proposed Wavelet–Hybrid CNN–LSTM model achieved 99% balanced accuracy and macro-F1 score in classifying materials from non-line-of-sight (NLOS) acoustic echoes, outperforming wavelet-only and classical acoustic models.

SHAP-based explainability analysis revealed that Mel-Frequency Cepstral Coefficient (MFCC) and wavelet entropy-energy features play complementary roles in material discrimination, allowing the model to not only classify with high accuracy but also interpret physical material properties such as hardness, density, and porosity.

What are the implications of the main findings?
The study demonstrates that hybrid wavelet–acoustic feature fusion can enable real-time, interpretable acoustic sensing systems for material recognition in NLOS environments such as robotics, defense, and industrial monitoring.The integration of deep recurrent models with interpretable hybrid features provides a foundation for developing physics-informed artificial intelligence systems, bridging the gap between data-driven learning and acoustic material physics.

The study demonstrates that hybrid wavelet–acoustic feature fusion can enable real-time, interpretable acoustic sensing systems for material recognition in NLOS environments such as robotics, defense, and industrial monitoring.

The integration of deep recurrent models with interpretable hybrid features provides a foundation for developing physics-informed artificial intelligence systems, bridging the gap between data-driven learning and acoustic material physics.

Acoustic material classification under non-line-of-sight (NLOS) conditions—where direct sound paths are obstructed—is a challenging task due to echo attenuation, complex reflections, and noise effects. This study aims to improve NLOS material recognition by introducing a novel wavelet–acoustic hybrid feature fusion method integrated with deep recurrent neural network architectures. Echo signals from nine different materials were collected using the newly developed ANLOS-R (Acoustic Non-Line-of-Sight Recognition) dataset, which was specifically designed to simulate realistic NLOS propagation environments. From these recordings, time-domain acoustic features and multi-scale wavelet-based energy and entropy statistics were extracted using ten wavelet families. The resulting 70-dimensional hybrid feature set was used to train several deep learning architectures, including Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), Gated Recurrent Unit (GRU), and Convolutional Neural Network–LSTM (CNN–LSTM). Among these, the CNN–LSTM achieved the highest balanced accuracy and macro-F1 score of 0.99, showing strong generalization and convergence performance. SHapley Additive exPlanations (SHAP) analysis indicated that Mel-Frequency Cepstral Coefficients (MFCCs) and wavelet entropy–energy features play complementary roles in material discrimination. The proposed approach provides a robust and interpretable framework for real-time NLOS acoustic sensing, bridging data-driven deep learning with the physical understanding of acoustic material behavior.