RESEARCH-ARTICLE
Underwater Ship Noise Recognition Integrating Gate Position Encoding and Multi-Scale Features
Executive Impact & Key Findings
This research introduces the Gated Hybrid Positional Encoding-Multi-Scale Feature Extraction Network (GHPE-MSFNet) to overcome limitations in traditional deep learning for underwater vessel noise identification, specifically addressing rigid positional encoding and insufficient multi-scale feature extraction. The GHPE-MSFNet model dynamically integrates positional information and combines a Multi-Scale Residual Module with fusion channel attention to enhance frequency domain feature extraction for complex underwater data. Evaluated on the DeepShip dataset, the model achieves 97.73% accuracy and 97.74% F1 score in four-class vessel recognition tasks, outperforming several mainstream deep learning models. These results demonstrate significant improvements in the reliability of vessel recognition systems for noisy underwater environments, aiding marine monitoring and management.
0%
Overall Accuracy
0%
F1 Score
0%
Precision
0%
Recall
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Understanding how the model dynamically adapts to positional information is crucial for robust underwater acoustic signal processing.
Gated Hybrid Position Encoding (GHPE) Mechanism
Compute Sine Encoding (PEsin)
→
Compute Learnable Encoding (PElearned)
→
Generate Dynamic Gating Weight (g = σ(α))
→
Fuse Encodings (PEhybrid)
→
Inject into Input Sequence (X + PEhybrid)
The Gated Hybrid Position Encoding (GHPE) module dynamically fuses sinusoidal and learnable positional encodings based on input signal patterns, adapting to time-varying underwater acoustics. This mechanism enables the model to adjust the fusion ratio based on input signal patterns dynamically, favoring sinusoidal encoding for strongly periodic low-frequency signals and learnable coding for highly time-varying high-frequency signals.
| Encoding Type | Precision | Recall | F1 Score | Accuracy |
|---|---|---|---|---|
| A1: Sinusoidal PE | 0.9713 | 0.9713 | 0.9713 | 0.9712 |
| A2: Rotary PE | 0.9728 | 0.9708 | 0.9717 | 0.9716 |
| A3: Randomized PE | 0.9641 | 0.9656 | 0.9632 | 0.9641 |
| GHPE (No Gating) + MSFNet | 0.9741 | 0.9733 | 0.9737 | 0.9735 |
| GHPE + MSFNet (Complete) | 0.9773 | 0.9772 | 0.9774 | 0.9773 |
The ablation study highlights the superior performance of GHPE compared to traditional and randomized position encoding methods, demonstrating its adaptability to the unique characteristics of underwater acoustic signals. The dynamic gating mechanism (GHPE No Gating vs GHPE Complete) shows a notable improvement.
Effective multi-scale feature extraction is vital for discriminating vessel noise across different frequency bands.
MS-SE-ResNet Feature Extraction Process
Parallel Convolutions (3x3, 5x5, 7x7)
→
Global Average Pooling for Channel Statistics
→
SE Module Channel Calibration (s = σ(W2. δ (W₁z)))
→
Concatenate & 1x1 Convolution for Yfusion
→
Weighted Residual Fusion with Main Branch
The Multi-Scale Residual Network Module with Channel Attention (MS-SE-ResNet) is designed to capture diverse frequency-domain features of ship noise. It employs parallel convolutions, followed by channel calibration via an SE module, and a weighted fusion mechanism.
| Module Type | Precision | Recall | F1 Score | Accuracy |
|---|---|---|---|---|
| B1: SE-ResNet50 based | 0.9559 | 0.9552 | 0.9554 | 0.9548 |
| B2: MSDANet based | 0.9487 | 0.9469 | 0.9477 | 0.9472 |
| B3: ConvNeXt based | 0.9651 | 0.9649 | 0.9650 | 0.9649 |
| GHPE-MSFNet (No SE) | 0.9683 | 0.9681 | 0.9681 | 0.9681 |
| GHPE-MSFNet (Complete) | 0.9773 | 0.9772 | 0.9774 | 0.9773 |
The MS-SE-ResNet module demonstrates superior performance compared to other multi-scale and attention-based architectures, highlighting its ability to capture complex scale interactions and extract discriminative features. The integration of SE attention (GHPE-MSFNet No SE vs GHPE-MSFNet Complete) proves crucial.
The comprehensive performance of the GHPE-MSFNet model for real-world applications.
0%
Overall Accuracy
The GHPE-MSFNet model achieved an outstanding 97.73% accuracy on the DeepShip dataset, demonstrating its robust performance for underwater ship noise recognition.
0%
Overall F1 Score
With an F1 score of 97.74%, the GHPE-MSFNet model showcases a strong balance between precision and recall, indicating its effectiveness in correctly identifying ship types.
Enhanced Robustness in Marine Monitoring
The GHPE-MSFNet model's superior adaptability to complex underwater data and its effective extraction of multi-scale, discriminative features lead to a highly reliable vessel recognition system. This robustness is critical for accurate marine monitoring and management, even in noisy underwater environments. The model demonstrated balanced stable performance across diverse ship categories (Tanker, Tug, Passenger Ship, and Cargo), validating its capability to handle complex and variable underwater acoustic conditions effectively. Future work suggests constructing richer datasets with complex environmental interferences (reverberation, multipath, biological noise, dynamic SNR) and exploring multimodal data fusion to further enhance applicability.
Projected ROI for Your Enterprise
Estimate the potential time savings and cost reductions by integrating advanced AI solutions into your operations.
Annual Cost Savings
$0
Annual Hours Reclaimed
0
Your AI Implementation Roadmap
A typical AI adoption journey, tailored to maximize efficiency and integrate seamlessly with your existing infrastructure.
Phase 1: Discovery & Strategy
Comprehensive analysis of existing workflows, identification of key AI integration points, and development of a custom implementation strategy aligned with business objectives.
Phase 2: Pilot & Development
Deployment of a small-scale pilot project to validate the AI solution, iterative development based on feedback, and initial model training with proprietary data.
Phase 3: Full-Scale Integration
Seamless integration of the AI solution across all identified departments, extensive testing, and staff training to ensure smooth adoption and maximum utility.
Phase 4: Optimization & Scaling
Continuous monitoring of AI performance, fine-tuning of models for enhanced accuracy and efficiency, and strategic planning for future AI expansions and innovations.
Ready to Transform Your Enterprise with AI?
Our experts are ready to discuss a tailored AI strategy that drives efficiency, innovation, and measurable ROI for your business.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat