AI meets Agriculture: A Smart System for Foreign Object Damage Avoidance

The Industry 4.0 trend is transforming the production capabilities of all industries, including the agricultural domain. This transition proves benefits when producing under variable conditions while still guaranteeing high productivity and quality. In agricultural applications, harvesters are becoming complex mechatronic systems equipped with a high number of sensors allowing to precisely monitor the machine and its surroundings where the machine is used. In this context, we published new detection algorithms with adaptive capabilities to detect foreign elements in a harvesting process based on knock sensor signals, to avoid damaging critical parts of the machine [9]. These methods already proved 95% correct detection when the data contains limited variability translated to limited noise floor and relatively clear transients due to foreign elements. In this study, we extend these detection methods to work with more variable noise floor and less clear transients in the signals, which represent realistic cases when harvesting is done in complex situations such as muddy crop, humid environment, etc. The new approach presents the detected transients as candidates to a classification method which decides whether the transient is due to a foreign element or not. Employing both a decision tree and random forest classifier, we obtained 100% correct detection of those transients in the signal that resulted from a foreign element.