Enhancing Survivor Localization Performance Using Residual U-Net for Interference Cancellation in Disaster Situations

This paper introduces a method to improve the accuracy of detecting and locating individuals in disaster situations with interference signals. The approach features an interference cancellation model that isolates and refines the rescue request signal, enhancing input quality for the localization model. The localization uses the generalized cross-correlation phase transform (GCC-PHAT) technique to estimate sound source positions. In interference-free conditions, the model achieves a mean absolute error (MAE) of 1.16 meters and a Euclidean distance (ED) error of 0.89 meters. However, in 0 dB SNR environments, the MAE worsens to 2.35 meters and the ED to 1.88 meters. With the interference cancellation model, the system improves, reducing the MAE to 1.81 meters and the ED to 1.43 meters under the same conditions. These results demonstrate the model's effectiveness in mitigating noise effects, enhancing survivor detection and localization accuracy, critical for timely rescue operations.