Unsupervised domain adaptation with self-training for weed segmentation
Bibliographic record
Abstract
Accurate crop and weed segmentation in varied field conditions is crucial for advancing automated weed management but remains challenging. Though promising, convolutional neural networks (CNNs) often experience performance drops when deployed in new field environments due to shifts between training and test data distributions. To address this limitation, we proposed a self-training framework using a teacher–student model that adapts CNNs for diverse agricultural contexts. Our method enhances generalization by co-training the student model on both the source domain and pseudo-labelled target domain generated by the teacher model, with teacher parameters updated via an exponential moving average of the student’s model. The main contributions of this work are as follows: (1) we simplified the self-training procedure by using all target predictions, skipping the selection phase, and applying local dynamic weights (LDW) for target pixels during co-training; (2) we optimized iteration by monitoring covariance fluctuations to avoid pseudo-label overfitting and reduced the impact of false labels; (3) we addressed class imbalance with dynamic class weights (DCW) to give more importance to minority classes; and (4) we formulated a loss function integrating both LDW and DCW into the soft intersection over union (softIoU), enhancing weed segmentation effectiveness. We evaluated our framework with the ROSE challenge dataset across eight adaptations involving varied plants, robots, and growth stages, achieving up to a 0.17 mean IoU improvement over popular methods like CycleGAN. Our approach demonstrated consistent performance across diverse agricultural environments, supporting its use in real-field inference. • We introduced a self-training framework for weed segmentation. • Four novel strategies were developed to enhance the performance. • The method was evaluated on datasets with varying domain gaps. • It is demonstrated generalization across different robots and growth stages. • The experimental code is publicly available for future research.
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 | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 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".