Alexander Sutin – asutin@stevens.edu
Alexander Yakubovskiy – ayakubov@stevens.edu
Hady Salloum – hsalloum@stevens.edu
Timothy Flynn – tflynn2@stevens.edu
Nikolay Sedunov – nsednov@stevens.edu
Stevens Institute of Technology, Hoboken, NJ 07030
Hannah Nadel – Hannah.Nadel@aphis.usda.gov
USDA APHIS PPQ S&T, 1398 West Truck Road, Buzzards Bay, MA 02542
Sindhu Krishnankutty – Sindhu.Krishnankutty@aphis.usda.gov
Department of Biology, Xavier University, Cincinnati, OH 45207
Popular version of paper 2pAB2 , “Sound of wood-boring larvae and its automated detection”
Presented Tuesday, May 8, 2018, 1:40-2:00 PM, LAKESHORE B,
175th ASA Meeting, Minneapolis.
Figure 1. Tree bolt with wood-boring beetle larva and attached sensors
The difficulty to detect potentially dangerous plant pests at ports of entry by agricultural inspectors, and the increasing invasion of U.S. agriculture and forestry by exotic pests in recent years are serious problems, given that the Federal Government instituted a robust reinforcement of the country’s borders and ports of entry. It is estimated that costs to the American economy caused by exotic invasive species are now over $138 B per year. Customs and Border Protection (CBP) facilitates processing of ~$2 trillion in legitimate trade, imports and exports yearly while enforcing U.S. trade laws that protect the economy, health, and the safety of people worldwide. Currently, CBP agriculture specialists inspect for pests relying on mostly manual techniques that are time-consuming and potentially not 100% effective because resources allow for only 2% of cargo to be examined. Wood boring pests are especially time-consuming to detect, as they burrow and feed inside wood, and often leave few visual cues to their presence.
Stevens Institute of Technology has been investigating engineering solutions to augment the current wood inspection process at ports of entry in an effort to minimize the time spent per inspection and maximize the detection rate of infestations in wood packaging and wood products. One of our systems is based on the detection of vibrational pulses make by wood boring larvae during feeding; results of the initial research in this direction are presented in [1]. A major problem of automated detection of wood-boring larvae is detecting insect-induced vibrational pulses with noise in the background. To develop an acoustic-signature detection algorithm, numerous acoustic signals made by the larvae of Anoplophora glabripennis (Asian Longhorn Beetle) and Agrilus planipennis (Emerald Ash Borer) were collected in a quarantine facility at the USDA-APHIS PPQ Otis laboratory. We also recorded and analyzed typical background noise pulses, namely, speech, knocking, and tapping made by humans, and sounds of electronic equipment. Examples of time tracks and spectrograms of the recorded signal are shown in Fig. 2. and 3.
Figure 2. Recorded vibrational pulses from various sources.
Figure 3. Spectrograms of insect sounds and human speech.
In the conducted analyses, we considered the features of both sound pulses of larvae and typical noise pulses. The extracted and evaluated features of those sound pulses were based on the estimation of the duration, spectrum and spectrogram of the signal, and spectrum and spectrogram of the signal envelope (estimated via Hilbert transform). Some noise pulses (knocking and tapping) are longer than the larval bite sounds, while some (electronic beeps) are similar in duration. Speech includes fragments (vowels) which are much longer than larval bite sounds, but also very short fragments inside the vowels (high-pitched harmonics). The spectral content of some non-insect sounds differ from larval feeding sounds. The envelope spectra, therefore, appear to be informative features.
Analysis of the recorded vibrations allowed extraction of signal features that could ultimately be used for larval classification. These features include the main frequency of the generated pulses, their duration, and main frequency of the pulse envelop (modulation frequency). In the conducted tests, these features show a clear separation of ALB and EAB acoustic signatures. For example, the main frequency of the ALB sound was in the range of 3.8-4.8 kHz, while for EAB it was between 1.2 and 1.8 kHz. A preliminary algorithm for automated insect-signal detection was developed. The algorithm automatically detects pulses with parameters typical for larva-induced sounds and rejects non-insect sound pulses that belong to the ambient noise. Detection is determined when the number of detected pulses for some time (1 min) exceeds the definite threshold. In the test, this algorithm detected a larva in all samples without false alarms.
We are close to the finalization of a prototype for wood-boring-insect detection in wooden pallets. This prototype includes the following features:
- Sensitive sensors that are practically unaffected by external sounds. These sensors contain an accelerometer and a microphone used for ambient noise estimation and elimination of strong ambient noise signals that can penetrate the vibrational channel.
- An insect sound emitter to simulate real insect sounds and apply it to the testing and calibration of the detection system.
- Detection of insect-produced vibrations, based on the principles presented above.
Acknowledgement
This project was funded under contract with the U.S Department of Homeland Security’s Science and Technology Directorate (S&T). The opinions contained herein are those of the contractors and do not necessarily reflect those of DHS S&T.
References
[1] Sutin, A., T. Flynn, H. Salloum, N. Sedunov, Y. Sinelnikov, and H. Hull-Sanders. 2017. Vibro-acoustic methods of insect detection in agricultural shipments and wood packing materials. In: Proceedings of Technologies for Homeland Security (HST), IEEE International Symposium, 2017, Boston, USA, pp. 1 – 6.
[embeddoc url=”https://acoustics.org/wp-content/uploads/2018/05/Sutin-LLP.docx” viewer=”microsoft”]