Detection for weak hydroacoustic signals of uwFBG marine sensors based on GA-PSO optimized PSR and SVM

This study addresses the detection challenge of weak hydroacoustic signals in low signal-to-noise ratio (SNR) marine environments by proposing a novel detection methodology for ultra-weak fiber Bragg grating hydrophones, integrating genetic algorithm-particle swarm optimization (GA-PSO)-optimized phase space reconstruction (PSR) with support vector machines (SVMs). A theoretical analysis is conducted on the weak signal processing capabilities of PSR and SVM algorithms. The parameters for both algorithms are optimized using PSO, with GA cross-mutation implemented to enhance global search capability and prevent premature convergence. Subsequently, we investigate the influence of key parameters, including time delay, embedding dimension, penalty factor, and kernel function width, on algorithmic error in weak signal processing. A comparative analysis is performed through controlled pool experiments to evaluate the performance of baseline SVM, PSO-SVM, and the proposed combined algorithm in processing both pulse and periodic weak signals. The experimental results show that GA-PSO-SVM joint processing reduced the pulse signal root mean square error (RMSE) from 0.003 580 to 0.000 108, while the periodic signal RMSE decreased from 0.001 400 to 0.000 157. Furthermore, the processed signals demonstrate SNR improvements exceeding 9 dB across frequency bands (500 Hz: 13.56 dB; 1000 Hz: 10.22 dB; 1500 Hz: 14.21 dB; 2000 Hz: 15.75 dB), with distinct directional patterns observed at 60° and 120° underwater orientations. The t-statistic results indicate that the mean difference of the GA-PSO-SVM algorithm is more significant than that of other traditional algorithms. The experimental results demonstrate that the proposed algorithm exhibits strong robustness for weak hydroacoustic signal processing in complex marine environments.