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

Qi Jiang, Yu Jie Liu, Yu Sun, Lili Ren, Youqing LuoInsects 2025 · DOI

TL;DR

This paper tackles the challenge of accurately identifying forest wood-boring insects, even when multiple species infest the same tree. They demonstrate that deep learning models, particularly ResNet, using acoustic spectrograms significantly outperform traditional machine learning methods in these complex co-infestation scenarios, achieving up to 88.75% accuracy. This work paves the way for more robust, real-time acoustic pest detection systems for forest management.

What Problem Does This Paper Address?

Forest wood-boring insects are a major threat to trees, often living hidden within the trunk. This makes early detection difficult, as damage symptoms only appear when the infestation is severe. Traditional detection methods have limitations, and current acoustic detection techniques struggle when multiple insect species infest the same tree simultaneously (co-infestation). These mixed acoustic signals can confuse recognition algorithms, leading to inaccurate identification and hindering effective pest management. This study aims to develop an acoustic detection system that can reliably identify wood-boring pests even in these challenging co-infestation scenarios.

Methods

The researchers employed two main approaches for acoustic signal analysis:

  • Machine Learning (ML) with Feature Variables: They extracted seven acoustic features (e.g., max amplitude, RMS amplitude, peak frequency) from the recorded insect sounds. Three ML models – Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN) – were trained and tested using these features.
  • Deep Learning (DL) with Spectrograms: They converted the acoustic signals into spectrograms, which represent the time-frequency distribution of the sound. Three DL models – AlexNet, ResNet, and VGG – were then used to classify these spectrograms.

The study designed three infestation scenarios: single-species infestation, co-infestation without mixed signals, and co-infestation with mixed signals, to rigorously test the models' performance.

Datasets

The study utilized two main datasets, derived from experimental logs of two tree species:

  • Dataset 1 (D1): Logs of Platycladus orientalis infested with either Semanotus bifasciatus or Phloeosinus aubei, or a combination of both.
  • Dataset 2 (D2): Logs of Fraxinus chinensis infested with either Agrilus planipennis or Streltzoviella insularis, or a combination of both.

For each dataset, subsets were created to represent the three infestation scenarios: single-species, co-infestation (without mixed signals), and co-infestation (with mixed signals). Background noise from non-infested logs was also included. A portion (80%) of the data was used for training, and the remaining 20% for testing.

Key Insights and Ideas

  • The challenge of mixed signals: The core difficulty lies in distinguishing individual insect sounds when multiple species are active simultaneously. The acoustic signatures can overlap and interfere, making it hard for algorithms to isolate and identify each species.
  • Spectrograms as richer representations: Spectrograms, which capture the time-frequency characteristics of sound, proved to be a more informative input for deep learning models compared to handcrafted acoustic features. This suggests that the complex patterns within spectrograms are crucial for disentangling mixed acoustic signals.
  • Deep learning's advantage in complexity: While traditional ML models excel in simpler scenarios (single species), their performance degrades significantly when faced with the complexity of mixed signals. Deep learning models, with their ability to learn hierarchical features directly from data, demonstrate a clear advantage in these challenging situations.
  • ResNet's superiority: Among the tested DL models, ResNet consistently showed the highest accuracy, indicating its effectiveness in handling the nuances of mixed acoustic signals from wood-boring insects.

Key Results

  • Single-species scenarios: All ML models achieved perfect accuracy (100% OA, Kappa=1) in identifying single pest species and distinguishing them from background noise.
  • Co-infestation without mixed signals: ML model accuracy slightly decreased but remained high. Dataset 1 (D1) generally showed better performance than Dataset 2 (D2), possibly due to more distinct acoustic differences between the pest species in D1.
  • Co-infestation with mixed signals: This is where the significant differences emerged. ML models' accuracy dropped substantially (below 75%), struggling to identify pests with weaker signals. In contrast, DL models showed a marked improvement.
  • In Dataset 1 (D1), DL models improved accuracy from 68.75–71.25% to 82.5–88.75%.
  • In Dataset 2 (D2), DL models improved accuracy from 63.75–72.5% to 77.5–85.0%.
  • ResNet's top performance: ResNet achieved the highest overall accuracy, reaching 88.75% in D1 and 85.0% in D2, effectively reducing interference from mixed signals.

Limitations and Future Work

  • Dataset limitations: The study acknowledges that the dataset size and diversity could be expanded. Future work will focus on including more pest species and host tree types to develop a more comprehensive general recognition model.
  • Signal attenuation: The impact of signal attenuation as vibrations travel through tree trunks is a critical factor for real-world deployment. Future research will explore these attenuation patterns to optimize sensor placement and improve detection accuracy.
  • Environmental noise: While background noise was considered, future efforts need to enhance the model's ability to discriminate against a wider range of environmental interference, including wind, rain, and other biotic noises.
  • Hardware development: The authors propose developing compact, low-power, IoT-oriented detection devices for field deployment, integrating signal amplification, data acquisition, and lightweight recognition models for real-time, on-site analysis.
  • Sensor technology: Future research will also compare their current acoustic sensor technology with distributed acoustic sensors (e.g., fiber-optic) to evaluate their performance and application potential.

Summary updated

Abstract

Acoustic detection technology has emerged as a promising, non-destructive and continuous monitoring method for pest early detection at the single tree level. However, field application still encounters problems, especially under complex infestation scenarios, i.e., co-infestations by multiple pest species. This study aims to develop a novel acoustic-based recognition model for detecting forest wood-boring pests, specially designed to enhance monitoring accuracy under complex infestation scenarios. We collected feeding vibration signals from four wood-boring pests: <i>Semanotus bifasciatus</i>, <i>Phloeosinus aubei</i>, <i>Agrilus planipennis</i>, and <i>Streltzoviella insularis</i>. Three infestation scenarios were designed: single-species, co-infestation without mixed signals, and co-infestation with mixed signals. Three machine learning (ML) models (Random Forest, Support Vector Machine, and Artificial Neural Network) based on seven acoustic feature variables, and three deep learning (DL) models (AlexNet, ResNet, and VGG) using spectrograms were employed to classify the signals. Results showed that ML models achieved perfect accuracy (OA: 100%, Kappa: 1) in single-species scenarios but declined significantly under co-infestation scenarios with mixed signals. In contrast, DL models, particularly ResNet, maintained high accuracy (OA: 85.0-88.75%) and effectively discriminated mixed signals. In conclusion, this study demonstrates the superiority of spectrogram-based DL models for acoustic detection under complex infestation scenarios and provides a foundation for developing a general, real-time detection model for integrated pest management in forest ecosystems.
BibTeX
@inproceedings{jiang2025acoustic,
  author = {Qi Jiang and Yu Jie Liu and Yu Sun and Lili Ren and Youqing Luo},
  title = {{Acoustic Detection of Forest Wood-Boring Insects Under Co-Infestations}},
  booktitle = {Insects},
  year = {2025},
  doi = {10.3390/insects16121241},
}
Figure