| Title |
Deep Learning-based Noise Reduction and Sound Classification for Rescue in Disaster Sites |
| Authors |
김준휘(Joonhwi Kim) ; 최준규(Jungyu Choi) ; 임성빈(Sungbin Im) |
| DOI |
https://doi.org/10.5573/ieie.2024.61.12.101 |
| Keywords |
Noise reduction; Disaster site; Wiener filtering |
| Abstract |
Urban disaster sites often experience various types of noise, which can significantly hinder the accuracy and efficiency of human detection and rescue operations. To address this issue, this study proposes a noise reduction method that combines a Wave-U-Net-based deep learning model with a Wiener filter. Clean audio signals and various types of noise were synthesized to create a dataset consisting of clean signals (data1), signals with noise reduction applied (data2), and signals with an additional Wiener filter applied (data3). The dataset was generated at noise levels ranging from SNR 0 dB to 30 dB in 5 dB increments, and each model was trained using these datasets. To evaluate the effectiveness of the proposed noise reduction method, the performance of Simple CNN (Convolutional Neural Network), XGBoost (eXtreme Gradient Boosting), and SVM (Support Vector Machine) was measured on each dataset. Experimental results showed that noise reduction had a positive effect on performance in the SNR 0 dB to 5 dB range, but in environments with SNR levels above 10 dB, noise reduction led to performance degradation. This degradation is likely due to over-suppression of noise in lower-noise environments, which can distort the signal or result in unnecessary signal processing that harms the original signal quality. This study demonstrates the effectiveness of noise reduction in complex noise environments and highlights the need for further research to develop methods that maintain performance in higher SNR environments, particularly above 10 dB. |