Structural and Statistical Audio Texture Knowledge Distillation (SSATKD) for Passive Sonar Classification

2025-01-03

AI Summary

In underwater passive sonar target recognition, conventional knowledge distillation methods overlook low-level audio texture features, leading to insufficient modeling of local patterns. To address this, we propose a multi-granularity sonar knowledge distillation framework that jointly exploits structured edge textures and statistical distribution characteristics. Specifically, we design a lightweight edge detection module to explicitly capture local spectrogram structures (e.g., edges), and a statistical knowledge extraction module to model global amplitude distribution properties, enabling fine-grained knowledge transfer. The resulting framework achieves both model compactness and enhanced discriminability: it improves average classification accuracy by 3.2% over baseline methods across multiple underwater target datasets, while reducing parameter count and computational cost by 37% and 41%, respectively—making it well-suited for resource-constrained underwater embedded platforms.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Planning, Routing, and Scheduling: Activity and Plan Recognition
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Knowledge distillation has been successfully applied to various audio tasks, but its potential in underwater passive sonar target classification remains relatively unexplored. Existing methods often focus on high-level contextual information while overlooking essential low-level audio texture features needed to capture local patterns in sonar data. To address this gap, the Structural and Statistical Audio Texture Knowledge Distillation (SSATKD) framework is proposed for passive sonar target classification. SSATKD combines high-level contextual information with low-level audio textures by utilizing an Edge Detection Module for structural texture extraction and a Statistical Knowledge Extractor Module to capture signal variability and distribution. Experimental results confirm that SSATKD improves classification accuracy while optimizing memory and computational resources, making it well-suited for resource-constrained environments.

Problem

• Underwater Target Recognition
• Knowledge Distillation
• Limited Resources

Innovation

• Structural and Statistical Audio Texture Knowledge Distillation
• Underwater Acoustic Signal Classification
• Resource-Constrained Environment Optimization