TY  - EJOU
AU  - Altayeva, Aigerim 
AU  - Omarov, Nurzhan 

TI  - A Lightweight Channel-Attention-Enhanced Deep Learning Architecture for Real-Time Hazardous Impulsive Sound Detection
T2  - Computers, Materials \& Continua

PY  - 2026
VL  - 89
IS  - 1
SN  - 1546-2226

AB  - Hazardous impulsive sound detection plays a critical role in intelligent surveillance, public safety monitoring, and automated emergency response systems. This study proposes a lightweight channel-attention-enhanced deep learning architecture for real-time detection and classification of hazardous acoustic events. The proposed framework utilizes mel-spectrogram representations to capture time-frequency characteristics of audio signals and employs a compact convolutional neural backbone to efficiently extract hierarchical features. To enhance feature discrimination, a squeeze-and-excitation channel-attention mechanism is integrated into the architecture, enabling adaptive recalibration of feature channels and improved robustness under noisy and complex acoustic environments. A custom dataset consisting of 8000 audio samples across eight hazardous sound classes was constructed to evaluate the proposed model. Experimental results demonstrate strong classification performance, achieving high accuracy, precision, recall, and F1-score while maintaining computational efficiency. The ablation study confirms the effectiveness of the channel-attention mechanism in improving classification performance. Furthermore, the lightweight design ensures suitability for real-time deployment on resource-constrained platforms. The proposed approach provides an efficient and reliable solution for hazardous impulsive sound detection and offers significant potential for integration into intelligent monitoring and safety-critical applications.
KW  - Hazardous sound detection; impulsive sound; deep learning; channel-attention mechanism; mel-spectrogram; real-time audio monitoring; lightweight neural networks

DO  - 10.32604/cmc.2026.080878