Automated acoustic detection of a cicadid pest in coffee plantations

Abstract South american countries are the largest coffee producers in the world. Nevertheless, Cicadidae, the colloquial term for cicadas, is one of the key pests responsible for dropping the production. Currently, there is no electronic device or autonomous technological resource commercially available for detecting certain species of cicadas in the crop, penalizing the farmers on the management of that insect. Thus, this article presents a novel algorithm implemented in a low-cost real-time plataform for the acoustic detection of cicadas in plantations. Based on the Bark Scale (BS), Wavelet-packet Transform (WPT), Paraconsistent Feature Engineering (PFE) and Support Vector Machines (SVMs), the proposed technique was assessed with a database of 1366 recordings, presenting a value of accuracy of 96.41% for the distinction among cicadas and background noise, where the latter includes sounds from mechanical devices, birds, animals in general and speech, among others.