From Signals to Evidence: Interpretable Underwater Acoustic Target Sensing with Large Language Models - Part II: Passive Detection

Zhiyuan Li; Yuge Zhang; Jie Chen; Xiruo Su; Zhengqiao Zhao; Xiling Yao; Lingji Xu; Jingdong Chen

cs.SDOct 22, 2026

From Signals to Evidence: Interpretable Underwater Acoustic Target Sensing with Large Language Models - Part II: Passive Detection

Authors: Zhiyuan Li, Yuge Zhang, Jie Chen, Xiruo Su, Zhengqiao Zhao, Xiling Yao, Lingji Xu, Jingdong Chen,

Organizations: School of Artificial Intelligence, Northwestern Polytechnical University, Xi’an, Shaanxi 710129, China · School of Ocean Engineering and Technology, Sun Yat-sen University, Zhuhai, 519000, Guangdong, China · School of Marine Science and Technology, Northwestern Polytechnical University, Xi’an, Shaanxi 710129, China · School of Electronic and Optical Engineering, Nanjing University of Science and Technology, Nanjing, Jiangsu 210094, China

Abstract

Underwater acoustic passive detection is essential to maritime surveillance and intelligent ocean sensing. Existing detection systems typically collapse multichannel array observations into a binary decision or a scalar confidence score, offering limited insight into the physical evidence that supports or contradicts a target declaration. This limitation is acute at low signal-to-noise ratio (SNR), where isolated spectral peaks and transient beamforming responses are inherently ambiguous and physically verifiable targets are characterized by an ensemble of coupled reinforcing cues. Building on the unified from-signals-to-evidence framework introduced in the companion Part I, this paper instantiates the framework for passive target detection. A physics-aware multichannel acoustic encoder encodes the array observations into complementary acoustic evidence tokens through a direction-guided architecture that combines minimum variance distortionless response (MVDR) processing with time-frequency and fine-frequency representations. These tokens are combined with environmental priors and fed into a pretrained large language model (LLM) that evaluates target-present and noise-only hypotheses and generates a structured detection report detailing target presence, bearing, spectral-line locations, received SNR, range, and per-field confidence scores. A physics-parameterized data engine extended to realistic oceanic channels provides traceable supervision at both discrete-label and structured-description levels. Experiments under white-noise and mixed ocean-background conditions demonstrate reliable detection, accurate bearing estimation, and physically interpretable structured reports, validating the feasibility of evidence-grounded underwater acoustic detection.

Figures & tables

Fig. 1 : Conceptual overview of the proposed UPD framework for interpretable underwater acoustic target detection. Part I: Framework & Recognition has been addressed in our previous work [ 1 ] .

Fig. 2 : Architecture of the proposed from-signals-to-evidence framework.

Fig. 3 : Task-oriented acoustic encoding and language alignment. The DOA, DET, and LINE branches share the same observation windows. Their internal Fourier analysis windows determine the spectral grids. DOA provides window-level directional queries to DET and recording-level directional keys and values to LINE. The three sequences and task summaries are projected into the language-model input space. A separate head reads frequency scores from the additional readout position.

Task	Sequence	Summary	Prediction	Supervision
Detection	$\mathbf{T}_{\mathrm{ac}}$	$\mathbf{c}_{\mathrm{det}}$	Target-presence probability	Binary cross-entropy
Bearing	$\mathbf{D}$	$\mathbf{c}_{\mathrm{doa}}$	Distribution over bearings	KL divergence with circular soft labels
Frequency	$\mathbf{F}$	$\mathbf{c}_{\mathrm{freq}}$	Independent scores over frequency bins	Region-balanced binary cross-entropy with hard negatives

TABLE I : Acoustic sequences, task summaries, and supervision.

Fig. 4 : Propagation environment and array geometry. (a) Simulated Bathymetry over the horizontal domain, with water depth in km. Open circles mark candidate target locations and the star marks the array center. (b) Receiver-array plan view, with coordinates in m. Filled circles indicate hydrophones and $a$ denotes array radius.

Parameter	Setting
Bathymetric grid	$201\times 201$ points; $0.5$ km spacing
Water sound speed	Constant, $1500$ m/s
Seabed properties	Sound speed $1700$ m/s; density $1.8$ g/cm 3 ; attenuation $0.5$ dB/wavelength
Propagation solver	$30$ – $830$ Hz at $40$ Hz spacing; $8000$ beams; launch angles $\pm 60^{\circ}$

TABLE II : Propagation Configuration

Partition	Source IDs	Target	Background	Total
TRAIN	480	2400	480	2880
DEV	96	96	96	192
CAL	96	0	288	288
TEST	96	1440	288	1728

TABLE III : Observation Counts for Each Background Dataset

Branch	Analysis window (s)	Analysis hop (s)	FFT points	Bin spacing (Hz)	Band (Hz)
DOA	0.5	0.25	1,000	2.0	40–700
DET	0.5	0.25	1,000	2.0	0–1,000
LINE	2.0	1.00	4,000	0.5	0–1,000

TABLE IV : Fourier analysis within the shared observation windows.

Background	$P_{\mathrm{d}}$ (%)	False alarms $N_{\mathrm{FP}}/N_{0}$	Bearing MAE ( ∘ )	Precision	Recall	F1
White noise	100.00	0/96	1.26	100.00	93.64	96.72
Mixed background	98.96	0/96	0.22	81.75	81.83	81.79

TABLE V : Encoder validation performance at

-30

dB.

SNR (dB)	$P_{\mathrm{d}}$ (%)	$P_{\mathrm{fa}}$ (%)	Bearing MAE ( ∘ )	Precision (%)	Recall (%)	F1 (%)	SNR MAE (dB)	Range MAE (km)
0	100.00	–	0.000	77.54	35.74	48.93	0.000	8.410
$-10$	100.00	–	0.000	65.74	38.25	48.36	1.979	8.222
$-20$	100.00	–	0.260	65.76	45.34	53.68	0.451	9.174
$-25$	100.00	–	0.260	64.81	51.84	57.61	0.625	9.462
$-30$	100.00	–	0.521	63.20	56.78	59.82	0.105	8.749
Overall	100.00	0.00	0.208	67.11	43.95	53.11	0.632	8.803

TABLE VI : System test performance under the mixed background.

Fig. 5 : Receiver inputs and structured-report examples. Both observations use the same receiver metadata and numerical bathymetry. The target-present case has reference bearing $60^{\circ}$ , horizontal range 25 km, and SNR 0 dB. Numerical fields and all 23 frequency candidates are saved predictions. The English explanation is an edited illustration of how a possible harmonic relation can be reported. The target-absent case returns null physical attributes and an empty frequency list. Support scores are truncated to two decimal places.

Underwater Acoustics

See all 23 papers

From Signals to Evidence: Interpretable Underwater Acoustic Target Sensing with Large Language Models - Part I: Framework and Recognition
Oct 15, 2026Yuge Zhang, Jie Chen, Zhengqiao Zhao +4Underwater Acoustics Audio Representation Learning
School of Artificial Intelligence, Northwestern Polytechnical University, Xi’an, Shaanxi 710129, China · School of Ocean Engineering and Technology, Sun Yat-sen University, Zhuhai, 519000, Guangdong, China · School of Marine Science and Technology, Northwestern Polytechnical University, Xi’an, Shaanxi 710129, China +1
UA-Bench: Benchmarking Large Audio-Language Models on Underwater Acoustic Recognition
Oct 12, 2026Zhe Sun, Liang An, Jiayu Yao +9Underwater Acoustics Audio-Language Model Evaluation
Southeast University
SonarLLM: A Native Sonar--Optical Multimodal Large Language Model for Underwater Perception
Aug 25, 2026Cong Su, longxuan ma, Ling Dong +4Multimodal Sensor Fusion Underwater Imaging
Faculty of Information Engineering and Automation, Kunming University of Science and Technology Kunming, China · Yunnan Key Laboratory of Artificial Intelligence Kunming, China

Oct 15, 2026J/K move · Enter open · S save

From Signals to Evidence: Interpretable Underwater Acoustic Target Sensing with Large Language Models - Part I: Framework and Recognition

Yuge Zhang, Jie Chen, Zhengqiao Zhao, Xiling Yao+3

School of Artificial Intelligence, Northwestern Polytechnical University, Xi’an, Shaanxi 710129, China · School of Ocean Engineering and Technology, Sun Yat-sen University, Zhuhai, 519000, Guangdong, China · School of Marine Science and Technology, Northwestern Polytechnical University, Xi’an, Shaanxi 710129, China+1

Underwater Acoustics Audio Representation Learning Physics-Informed ML Audio-Language Models

Open paper AskPDF arXiv

Underwater acoustic target detection and recognition are essential to marine surveillance and intelligent ocean sensing. Despite substantial advances in deep learning, existing systems typically produce binary detection decisions, categorical labels, or scalar confidence scores, offering limited insight into the physical evidence underlying their outputs. Large language models (LLMs) offer an opportunity to move underwater acoustic sensing toward interpretable, evidence-grounded understanding; however, generic audio encoders struggle to capture the distinctive physical characteristics of underwater acoustics, and physically annotated data are scarce. In this paper, the first of a two-part series, we propose a from-signals-to-evidence framework that integrates physics-aware acoustic encoding with a pretrained LLM, organized around a unified feature-evidence-decision chain with a shared reasoning core. An open input layer accommodates heterogeneous information-raw waveforms, time-frequency representations, environmental and geographic data, and textual physical priors-unifying detection and recognition within a common formulation. In Part I, we instantiate this framework for target recognition through an Underwater Acoustic Physics-aware Transformer (UPT), which tokenizes complementary acoustic representations and learns task-disentangled physical representations through four physics-oriented learning tasks. We develop a physics-parameterized data engine that provides traceable supervision at two levels of granularity and introduce an attribute-level evaluation protocol to assess the physical fidelity of generated descriptions. Experiments on simulated recordings of four vessels demonstrate accurate vessel recognition and reliable estimation of shaft frequency, blade-passing frequency, and machinery-related attributes.

Abstract

Underwater Acoustics

Explore similar work

From Signals to Evidence: Interpretable Underwater Acoustic Target Sensing with Large Language Models - Part I: Framework and Recognition

UA-Bench: Benchmarking Large Audio-Language Models on Underwater Acoustic Recognition

SonarLLM: A Native Sonar--Optical Multimodal Large Language Model for Underwater Perception