Neural Network Optimization for Sonar Target Classification

In recent years, the application of neural networks to sonar target classification has advanced through improvements in both model architectures and optimisation strategies. Traditional gradient-based methods have delivered high accuracies but often struggle with local minima and heavy computational demands in underwater contexts characterised by noise and reverberation. To overcome these challenges, researchers have integrated metaheuristic algorithms, such as particle swarm and grasshopper optimisation, enabling robust feature selection and hyperparameter tuning for both shallow and deep networks. Concurrently, the exploitation of micro-Doppler signatures and bespoke convolutional and recurrent architectures has facilitated finer discrimination between acoustic returns of diverse targets. These innovations support real-time deployment on embedded platforms and enhance performance in defence, marine biology and environmental monitoring. The global significance of these techniques is evidenced by their potential to improve situational awareness in naval operations, automate marine species identification and monitor maritime traffic with minimal human supervision. Ongoing work seeks to unite resource-efficient training, adaptive model tuning and domain-specific signal processing to achieve resilient and generalisable sonar classifiers.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Particle Swarm Optimisation (PSO) has been applied to select discriminative features in micro-Doppler signatures, producing near-perfect recognition rates when coupled with k-nearest neighbour classifiers on both simulated and experimental datasets. An ensemble framework weighted by PSO further enhances reliability by calibrating contributions from multiple base learners applied to micro-Doppler data. The Grasshopper Optimisation Algorithm (GOA) has been used to train multilayer perceptron networks alongside simultaneous feature selection on high-dimensional sonar datasets, achieving classification accuracies above 98 per cent with rapid convergence and avoidance of local traps. More recently, deep architectures combining convolutional neural networks and long short-term memory layers have benefited from fuzzy slime mould optimisers for hyperparameter search, leading to lower false-alarm rates, faster convergence milestones and improved generalisation in sequential sonar sound classification.

Neural Network Optimization for Sonar Target Classification publication trend

The graph below shows the total number of articles in neural network optimization for sonar target classification across all publications each year (not limited to Nature Index journals).

Technical terms

Sonar target classification: The process of identifying and categorising underwater objects by analysing reflected acoustic signals.

Micro-Doppler effect: Variations in Doppler frequency induced by moving parts of a target, used to extract characteristic motion features.

Hyperparameter: A configuration setting of a neural network or optimisation algorithm determined prior to the training phase.

Metaheuristic optimisation: A class of problem-solving methods inspired by natural processes, designed to find near-optimal solutions for complex search spaces.

Particle Swarm Optimisation (PSO): A population-based algorithm that emulates social behaviour of flocks to iteratively improve candidate solutions.

Grasshopper Optimisation Algorithm (GOA): A nature-inspired method modelling the swarming behaviour of grasshoppers to perform global search in optimisation tasks.

Convolutional Neural Network (CNN): A deep learning model using convolutional layers to automatically learn spatial hierarchies of features from input data.

Long Short-Term Memory (LSTM) network: A recurrent neural architecture capable of capturing long-range temporal dependencies in sequential data.

References

1. Automatic recognition of sonar targets using feature selection in micro-Doppler signature. Defence Technology (2023).
2. Decision Fusion and Micro‐Doppler Effects in Moving Sonar Target Recognition. International Journal of Intelligent Systems (2023).
3. Feature Selection and Training Multilayer Perceptron Neural Networks Using Grasshopper Optimization Algorithm for Design Optimal Classifier of Big Data Sonar. Journal of Sensors (2022).
4. FUZ-SMO: A fuzzy slime mould optimizer for mitigating false alarm rates in the classification of underwater datasets using deep convolutional neural networks. Heliyon (2024).