科研之友
微信
新浪微博
Facebook
分享链接
摘要
Stink bugs cost the southeastern US cotton industry millions of dollars each year in crop losses and control costs. These losses are reduced by strategic pesticide applications; however, current methods of monitoring these pests for making management decisions are time-consuming and costly. Therefore, improved methods to identify and monitor these bugs must be investigated in order to optimize pesticide applications. One such method would be to exploit the substrate-borne vibrational signals (SBVSs) of these insects. Recordings of SBVS for two prevalent regional pests, the brown stink bug, Euschistus servus, and southern green stink bug, Nezara viridula, were segmented into separate pulses of variable duration based on signal energy. For each pulse, the linear frequency cepstral coefficients, dominant frequency, and duration were calculated and used as features. These features were classified using a Gaussian mixture model (GMM) and a probabilistic neural network (PNN) to discriminate these SBVS from incidental sounds and SBVS of different species from each other. Detection of SBVS generated by brown stink bugs was performed with over 92% accuracy for single male-female pairs with both PNN and GMM and with over 86% accuracy for 30 individuals with both PNN and GMM. Detection of SBVS generated by southern green stink bugs was performed with up to 82.5% accuracy with PNN and 68.0% accuracy with GMM for 30 individuals. Also, both PNN and GMM were over 90% accurate in identifying SBVS of brown and southern green stink bugs. Concurrent detection of SBVS from noise and identification of SBVS of brown and southern green stink bugs was 83.3% accurate using PNN and 71.5% accurate using GMM. These results indicated the capability of detecting and identifying stink bug species using their SBVS. Published by Elsevier B.V.
- 出版日期2013-2
