Classification of auditory signals from a combine harvester based on Mel-frequency Cepstral coefficients and machine learning

[摘要] As agricultural machinery moves into the digital era,significant developments in available technology will likelymake autonomous farm vehicles more feasible, affordable,and desirable. One of the challenges of effectiveautonomous vehicle control specific to agriculture is theability of the vehicle to interpret and adapt to constantlychanging conditions. Auditory information is a primaryindicator of changing conditions to an in-cab operator,particularly in situations such as detecting mechanicaloverload in a combine. This paper explores the potential forauditory information to be used in autonomous vehiclecontrol. The sound was recorded at a sampling rate of 48kHz near the straw chopper of a combine for three differentoperating modes during the same harvest day. Samples fromeach clip were segmented and analyzed to extract 31 audiofeatures. Six different feature selection methods ranked theimportance of each of the 31 features to identify the featuresthat lead to accurate classification with a minimal numberof calculations. These six rankings were assessed by Fagin’salgorithm to yield two features (both mel-frequency cepstralcoefficients). Twenty-five distinct machine learningclassification methods were evaluated using these twofeatures. Three of these classification methods reached100% accuracy, and 9 classifiers exceeded an individualsuccess rate of more than 99% using those same features.These feature extraction and classification steps took lessthan 1 s, assuring that such a classification system could beimplemented in real-time.

[学科分类] 农业科学（综合）

[关键词] Autonomous machines;remote supervision;auditoryinformation;machine monitoring;mel-frequency cepstralfeatures;machine learning;feature selection;Fagin’salgorithm