earticle

영어

Noise at construction sites is a major cause of hearing loss and mental damage for construction workers. In order to ensure noise safety, a new technology is needed to classify construction noise that affects worker’s safety. This paper proposes a machine learning-based construction equipment noise classification technique. In this study, three feature domains including time, frequency, and MFCC(Mel-Frequency Cepstral Coefficient) features were extracted from the audio data, and four different construction equipment were classified using three classifiers. For validation, a series of laboratory experiments are conducted. From the test, two classifiers that are KNN(K-Nearest Neighbor) and SVM(Support Vector Machine) show a high classification accuracy compared to DTs(Decision Trees) classifier. In addition, among the three features domains, MFCC features are found to be the most effective one in classifying the four construction equipments. Moreover, in order to investigate the reason for the classification accuracy difference, overlapping feature data between different equipments are analyzed in the feature domains.