PAFTI-Net predicts termite infestation risk levels from multivariate gas monitoring data
In response to the problems that traditional termite monitoring is difficult to quantify the degree of invasion and that there are noise interference in multi-temporal data as well as difficulties in jointly modeling local–global patterns, this paper proposes and implements Parallel Adaptive Fusion and Temporal Interaction Network (PAFTI-Net) based on the self-built gas monitoring platform. The model’s front end adopts adaptive feature fusion to dynamically weight CO2, CH4, temperature and humidity, and temporal features; the middle layer constructs three parallel paths: multi-scale convolution, cross-temporal global dependency modeling, and factor-level nonlinear interaction; the back end outputs low, medium, and high risk levels through dynamic fusion. On a structured dataset containing 89 independent experimental batches and 26,820 window samples, the test accuracy of PAFTI-Net reaches 91.50%, and the F1-score is 91.03%. Compared with the optimal baseline model ESD-TripleStream, it has achieved absolute improvements of 2.95% and 2.40% respectively, and a 3.69% increase in recall rate. The results indicate that PAFTI-Net has significant advantages in discriminating termite invasion risk levels under controlled conditions, providing a feasible technical path for intelligent prevention and control based on gas monitoring.
Authors
- Weiling Lu
- Shan Wu
- Hangjun Wang
Institutions
- Zhejiang A & F University (CN)
- Jiyang College of Zhejiang A&F University (CN)
- Huzhou College (CN)
Publication Details
- Journal
- Scientific Reports
- Published
- 2026-11-20
- DOI
- https://doi.org/10.1038/s41598-026-81080-1
- Primary Topic
- Smart Agriculture and AI
- Type
- article
- Field-Weighted Citation Impact
- 0.00