Acoustic detection and automatic identification of insect stages activity in grain bulks by noise spectra processing through classification algorithms.

The activity of insects within a grain bulk produces noises in the audible range of wavelengths, which can be detected by high performance acoustic sensors. A portable probe of 1.4 m length was built up with three levels acoustical sensors coupled to a computer-assisted processing system. The recorded sound signals of the major grain insect species were digitized and stored into a reference database. A classification algorithm was developed for the automatic recognition of recorded insect noise signals by their comparison to the specific spectra of the reference database. The system was calibrated for Sitophilus oryzae and Rhyzopertha dominica , either at the adult or larval stage. The performances of the computer-assisted acoustic probe have been checked in small pilot scale conditions of 300-kg grain bin units. The distance to the sensor from which the insect noise spectrum was no more accurately identifiable was assessed at 20 cm, representing a “sampled volume” equivalent to 65 kg of wheat gain at each probing. For S. oryzae in wheat grain, the relationship between insect activity (either of larval or adult stage) and density levels was quantitatively modelled in the range from one individual per 10 kg to 10 individuals per kg at temperature levels from 5 to 30°C. The cold stupor temperature of the rice weevil larva was assessed through the determination of the temperature level at the complete stop of activity deduced from the acoustical data. The threshold of temperature enabling insect larva activity was observed as low as 8°C, i.e. much lower than was previously established, published, or believed. This new tool of early detection of an infestation in grain bulk will be now associated to decision support system for IPM implementation in grain handling and storing plants.