Recognition of the Duration and Prediction of Insect Prevalence of Stored Rough Rice Infested by the Red Flour Beetle (Tribolium castaneum Herbst) Using an Electronic Nose

Sai Xu, Zhiyan Zhou, Keliang Li, Sierra Mari Jamir, Xiwen LuoSensors 2017 · DOI

TL;DR

This paper demonstrates the feasibility of using an electronic nose to monitor stored rough rice for red flour beetle infestations. By analyzing the volatile organic compounds emitted by infested rice, the system can accurately predict the duration and prevalence of insect infestation, offering a promising non-destructive method for grain quality control.

What Problem Does This Paper Address?

Ensuring the quality and safety of stored grains is crucial, especially in regions heavily reliant on rice like China. A major threat to stored grains is infestation by pests, such as the red flour beetle (Tribolium castaneum). Traditional methods for detecting these insects, like manual inspection, are labor-intensive, time-consuming, and often not sensitive enough for early detection. While other methods like acoustic or image analysis exist, they have limitations in real-world storage environments. This paper addresses the need for an intelligent, non-destructive, and efficient method for monitoring insect infestation in stored grains, specifically focusing on identifying both the duration of the infestation and the prevalence (amount) of insects.

Methods

The researchers employed an electronic nose (PEN3 AIRSENSE) equipped with a sensor array to detect volatile organic compounds (VOCs) released by rough rice samples. They created four experimental groups: one control (no infestation) and three infestation levels (light, middle, and heavy) of red flour beetles. Samples were collected at various time points after infestation to track the progression.

To analyze the complex data generated by the electronic nose, several pattern recognition techniques were applied:

  • Linear Discriminant Analysis (LDA): Used for dimensionality reduction and classifying infestation durations.
  • K-means Clustering Analysis (K-means) and Fuzzy C-means Analysis (FCM): Employed for clustering the data to recognize different infestation durations. FCM was found to be more robust than K-means.
  • Back-Propagation Artificial Neural Network (BPNN): Utilized for predicting the actual number of live insects (prevalence) in the infested rough rice.

Data pre-processing involved a reference signal removal method to mitigate drift noise common in time-series electronic nose data.

Datasets

The study used "Meixiangzhan" variety rough rice harvested in December 2014. The rice was naturally dried to a stable water content of 12-14%. For the experiment, 60g samples of rough rice were placed in 200 mL glass beakers.

  • Control (NT): No infestation.
  • Light Degree (LD): Infested with 10 red flour beetles.
  • Middle Degree (MD): Infested with 50 red flour beetles.
  • Heavy Degree (HD): Infested with 100 red flour beetles.

Samples were collected at -2 days (pre-infestation) and 2, 7, 13, 22, 28, and 36 days post-infestation. This resulted in a total of 140 electronic nose samples (4 degrees × 5 replications × 7 durations). For BPNN prediction, a subset of data was used for training (84 samples) and testing (21 samples).

Key Insights and Ideas

  • Volatile Signatures of Infestation: The study confirms that insect infestation, even at different degrees and durations, produces distinct volatile organic compound profiles that an electronic nose can detect.
  • Natural Insect Decline: The research observed a natural decline in live insect numbers over time, with the rate of decline being faster for higher infestation degrees. This decline slows significantly when insect populations drop below a certain threshold (around 13 individuals/60g), indicating a potential window for manual intervention.
  • FCM Superiority for Duration: Fuzzy C-means clustering proved more effective than K-means for recognizing infestation durations, likely due to its ability to handle overlapping data and avoid local optima.
  • BPNN for Prevalence Prediction: The BPNN model demonstrated a strong capability to predict the actual number of live insects, suggesting the electronic nose can quantify infestation levels.
  • Early Infestation Detection Advantage: The electronic nose method showed the best predictive performance when the infestation was just beginning (LD), highlighting its potential for early warning systems.

Key Results

  • LDA: Performed well for MD and HD duration recognition (accuracy >86%), but was poorer for LD.
  • K-means: Achieved accuracies of 88.57% (LD), 91.42% (MD), and 91.42% (HD) for duration recognition.
  • FCM: Significantly outperformed K-means, with accuracies of 94.29% (LD), 100% (MD), and 100% (HD) for duration recognition.
  • BPNN: Showed good predictive ability for insect prevalence, with determination coefficients (R²) for test sets ranging from 0.8912 (MD) to 0.9776 (LD). The predictive power was best for LD, intermediate for MD, and worst for HD.

Limitations and Future Work

  • LD Recognition Difficulty: The electronic nose's ability to distinguish infestation durations was less effective for the light infestation (LD) degree. This is attributed to the weaker and slower-changing volatile signals from less infested rice.
  • Interference in High Infestation: For heavily infested samples (HD), the BPNN prediction accuracy was lower. The authors suggest this might be due to increased interference from the smells of dead insects and the metabolism of live insects in a confined space.
  • Future Directions: The authors propose further research into overcoming interference issues, especially in heavily infested scenarios. They also suggest combining electronic nose detection with BPNN models for practical, large-scale granary monitoring through random sampling. The potential for using this technology for early detection of insect infestation is a key takeaway for future applications.

Summary updated

BibTeX
@article{xu2017recognition,
  author = {Sai Xu and Zhiyan Zhou and Keliang Li and Sierra Mari Jamir and Xiwen Luo},
  title = {{Recognition of the Duration and Prediction of Insect Prevalence of Stored Rough Rice Infested by the Red Flour Beetle (Tribolium castaneum Herbst) Using an Electronic Nose}},
  journal = {Sensors},
  year = {2017},
  doi = {10.3390/s17040688},
}
Figure