A Waveform Mapping-Based Approach for Enhancement of Trunk Borers' Vibration Signals Using Deep Learning Model.

Area of Science:

• Agricultural Entomology
• Bioacoustics
• Machine Learning

Background:

• Trunk-boring beetle larvae are difficult to detect as they cause minimal external damage.
• Vibration analysis is a promising method for detecting internal larvae activity.
• Environmental noise significantly hinders the accuracy of vibration-based detection systems.

Purpose of the Study:

• To develop a novel deep learning model for enhancing boring vibration signals.
• To improve the accuracy of detecting trunk-boring beetle larvae using acoustic methods.
• To address the challenge of noise interference in bioacoustic monitoring.

Main Methods:

• A deep learning model, VibDenoiser, was constructed using convolutional encoder-decoder architecture with SRU++ layers.
• The model was trained on a dataset comprising boring vibrations of *Agrilus planipennis* and environmental noise.
• The model leverages speech enhancement techniques for signal processing.

Main Results:

• VibDenoiser achieved a significant improvement of 18.57 dB in Signal-to-Noise Ratio (SNR).
• The model operates in real-time on a standard laptop CPU.
• Classification accuracy of four different models increased substantially when using enhanced vibration data.

Conclusions:

• The VibDenoiser model demonstrates high performance in enhancing noisy boring vibration signals.
• This work contributes a valuable tool for improving the detection of trunk-boring beetles.
• The developed model has practical implications for pest management and forest health monitoring.