Linear prediction of one-sided autocorrelation sequence for noisy acoustics recognition of excavation equipments

Abstract

Underground pipeline network suffers severely destructions by external excavation equipments in most of the developing countries in nowadays. Thus, it is urgent to construct an intelligent surveillance system to automatically detect earthmoving excavations along the pipeline network. In this paper, we analyze the acoustic signal based recognition method for excavation equipment detections. To enhance the recognition performance, a new feature of acoustic signals based on the AR model named the one-sided autocorrelation linear predictive cepstrum coefficients (OSALPCC) is employed for excavation equipment representation. Compared to the linear predictive cepstrum coefficients (LPCC) feature extraction approach, OSALPCC is robust to noises. Experiments on real acoustic signals for four representative excavation devices collected on a real metro construction site are conducted to show the effectiveness of the proposed method. Two intelligent algorithms, support vector machine (SVM) and k-nearest neighborhood (kNN), are adopted to test the recognition performance.