Agricultural Harvester Sound Classification using Convolutional Neural Networks and Spectrograms

Highlights Automatic classification of harvester sounds. Final obtained using three convolutional neural networks. The results the networks were combined via stacking and voting to achieve 100% accuracy. Abstract. use deep learning in agricultural tasks has recently become popular. Deep have been used for analyzing images crops, identifying paddy areas, distinguishing sick plants from healthy ones, name a few applications. Besides visual systems, sound analysis machinery is time-sensitive task that can also be incorporated decision making done with help models. We propose method generate spectrogram classify them into working modes real-time. outputs these as inputs ensemble improve accuracy system. To accuracy, final made by based on several consecutive classifications step. able perform less than 1 s which was standard considered safe time harvester. Keywords: Convolutional networks, learning, Spectrograms, Stacking, Voting.