LE-PWDNet: a lightweight and enhanced detection framework based on DEIM for early-stage pine wilt disease
Bibliographic record
Abstract
Pine wilt disease (PWD), characterized by rapid transmission and high pathogenicity, causes severe ecological and economic damage worldwide. Early detection is critical for curbing its spread, yet the concealed symptoms and minute lesions make it difficult for existing models to balance high accuracy with lightweight efficiency in complex forest environments. To address these challenges, this study proposes a lightweight detection model named LE-PWDNet. A total of 41,568 high-resolution UAV images were collected from diverse field scenarios to construct a standardized dataset covering four infection stages, providing comprehensive support for model training and evaluation. The model is built upon the DEIM training paradigm to enhance the utilization of positive samples for small-target detection. To strengthen multi-scale texture modeling of early lesions, a Wavelet Detail Attention Convolution (WDAConv) is designed. A ConvFFN module is introduced to mitigate the attenuation of high-frequency details, thereby improving robustness under complex backgrounds. A CGAFusion module is developed to reduce false positives caused by background noise. Furthermore, an Edge-Dilated Sampling-Point Generator (DySample-E) is incorporated to dynamically adjust the upsampling process, enhancing the ability to capture early micro-lesions. Experimental results demonstrate that, with only 5.64M parameters and approximately 7 GFLOPs, LE-PWDNet achieves an AP 50 of 83.8% for early-stage lesion detection and an overall AP 50 of 90.2%, outperforming existing mainstream models. This study provides a feasible solution for building intelligent and low-cost early-warning systems for forest diseases and highlights the broad application potential of the proposed framework in forestry and other ecological monitoring scenarios.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".