Deep Belief Network based audio classification for construction sites monitoring

• High-performance of deep belief networks for environmental sound classification. Audio-based classification construction equipment and tools. unmanned monitoring sites. In this paper, we propose a Deep Belief Network (DBN) based approach the audio signals to improve work activity identification remote surveillance projects. The aim is obtain an accurate flexible tool consistently executing managing sites by using distributed acoustic sensors. ten classes multiple tools, frequently broadly used in sites, have been collected examined conduct validate proposed approach. input provided DBN consists concatenation several statistics evaluated set spectral features, like MFCCs mel-scaled spectrogram. architecture, along with preprocessing feature extraction steps, has described details while effectiveness idea demonstrated some numerical results, real-world recordings. final overall accuracy on test up 98% significantly improved performance compared other state-of-the-are approaches. A practical real-time application presented method also order apply scheme data recorded different scenarios.