Acoustic Detection of Forest Wood-Boring Insects Under Co-Infestations.

Area of Science:

• Forestry
• Agricultural Entomology
• Acoustic Signal Processing

Background:

• Acoustic detection offers non-destructive, continuous pest monitoring at the single tree level.
• Field application faces challenges, particularly with co-infestations by multiple pest species.

Purpose of the Study:

• Develop a novel acoustic-based recognition model for forest wood-boring pests.
• Enhance monitoring accuracy in complex infestation scenarios, including co-infestations.

Main Methods:

• Collected feeding vibration signals from four wood-boring pest species.
• Designed single-species and co-infestation (with/without mixed signals) scenarios.
• Employed machine learning (Random Forest, SVM, ANN) and deep learning (AlexNet, ResNet, VGG) models for signal classification.

Main Results:

• Machine learning models achieved 100% accuracy in single-species scenarios but declined significantly with mixed signals.
• Deep learning models, particularly ResNet, maintained high accuracy (85.0-88.75%) and effectively discriminated mixed signals.
• Spectrogram-based deep learning models demonstrated superiority in complex infestation scenarios.

Conclusions:

• Deep learning models using spectrograms are superior for acoustic pest detection in complex forest infestations.
• This research provides a foundation for real-time, integrated pest management systems in forest ecosystems.