AUTOMATIC BIOACOUSTIC DETECTION OF RHYNCHOPHORUS FERRUGINEUS ‡ Ilyas Potamitis, † Todor Ganchev, † Nikos Fakotakis ‡ Department of Music Technology and Acoustics, Technological Educational Institute of Crete, Daskalaki-Perivolia, 74100, Rethymno, Greece, potamitis@stef.teicrete.gr † Department of Electrical and Computer Engineering, University of Patras, 26500 Rion-Patras, Greece, {tganchev, fakotaki}@wcl.ee.upatras.gr ABSTRACT The goal of this project is to research and develop a bio- acoustic automatic detector of a devastating pest attacking palm trees, which has recently appeared in Mediterranean countries. The method is based on piezoelectric sensors that are inserted in the tree trunk and record locomotion and feeding vibrations of the insect. The obtained signals are amplified, filtered, parameterized and automatically classified by advanced machine learning methods on a portable computer. We report excellent detection results reaching 99.5% on real-field recordings. 1. INTRODUCTION The world of insects is interesting from both a scientific perspective, for investigating the behaviour and diversity of biological organisms, and a practical perspective, due to fi- nancial reasons, as insects act as beneficial organisms in agriculture and forestry (they play a significant role in the food chain of other species and the fertility of plants). However, a number of insect species also have a negative bearing on agricultural economy as they are threat to plants and crops. Insects are mainly categorized according to their ap- pearance [1-2] and sound production [3-4] that are species- specific. The localization and species recognition of insects are usually carried out manually, using trapping and obser- vation methods. The detection and identification process is in most cases a highly complex procedure because insects are heard more often than seen or trapped (especially those that live in complex environments or demonstrate nocturnal activity). In brief, acoustic identification of insects is based on their ability to generate sound either deliberately as a means of communication or as a by-product of eating, flying or locomotion. Provided that the bioacoustic signal produced by insects follows a consistent acoustical pattern that is spe- cies-specific, it can be employed for detection and identifi- cation purposes. Despite some rise of interest in the recent years [1-5], the applicability of machine learning tech- niques to the insect identification problem is still in its in- fancy. In this paper we present ongoing research on detection and early warning of pests dangerous to agriculture. Par- ticularly we focus on a devastating pest for palm planta- tions, namely Rhynchophorus ferrugineus (Red Palm Wee- vil -- RPW). RPW has been detected in several Mediterra- nean countries (Spain, Egypt, Greece, Cyprus and Israel). Its presence is confirmed by the official authorities of more than 30 Asian and African countries. Red Palm Weevil (RPW), is the most dangerous and deadly pest to date for coconut, oil, date, sago and other palms. Wherever it has been located its presence caused severe catastrophes in the plantation of palms. RPW has been detected in the islands of Crete and Rhodes in Greece and in Cyprus. Collected specimens of the weevil have been sent to the Museum of Physical His- tory of London, United Kingdom where their identity has been verified officially (Benakio Phytopathological Insti- tute, November 2005 internal communication). The cause of the high rate of spread of this pest is human intervention, by transporting infested young or adult date palm trees and offshoots from contaminated to uninfected areas. Date palm is an important crop in North African and Asian countries and ornamental palms are widely planted as amenity trees in the whole Mediterranean area. The pest attacks palm trees and if left undetected within a few weevil generations results in very severe overall damage in the plantation. The main problem is that the attack by the weevil is visible only when the tree has been fatally wounded and adult insects have escaped and infested other trees (RPW has a strong flying capability). Moreover the treatment with insecticides is not efficient since the trunk protects the pest. If the pest is detected on time, the damage can be minimized. The in- fested trees once located are destroyed, endangered trees are treated and biological traps are deployed. This treatment saves the rest of the plantation from being infested. The main thrust of this work is the development of a system for automatic acoustic recognition of the RPW acoustic signal, by employing suitable piezoelectric probes with uncoated waveguides that are inserted in the tree trunk and supervised recognition algorithms that classify re- corded sound vibrations. In the present contribution, we minimize human intervention by exploiting machine learn- ing techniques and a signal parameterization method, which was explicitly designed for the needs of insect recognition. 16th European Signal Processing Conference (EUSIPCO 2008), Lausanne, Switzerland, August 25-29, 2008, copyright by EURASIP