Table 1 Human-in-the-loop (HITL) framework used in this study
From: Evaluating Broadscale Deep Learning for Maya Settlement Detection in G-LiHT Lidar
Stage | Automated process | Human involvement |
|---|---|---|
1. Data Collection | Aggregation of Lidar datasets from multiple regions | Selection of representative sites; evaluation of coverage quality |
2. Preprocessing | Noise filtering, Lidar visualization, feature extraction | Review of visualization output; feedback on terrain and structure clarity |
3. Sample Labeling | Initial automated proposals using previous models | Manual annotation, correction and confirmation of sample labels |
4. Model Training | Deep learning model refinement on labeled samples | Regular review of intermediate results; adjustment of training parameters |
5. Model Evaluation | Accuracy, precision, recall, F1 score metrics | Review of object identifications; input on evaluation methods, cultural relevance |
6. Application and Discovery | Deployment on new regions for predictive mapping | Final review of model outputs; visual inspection and expert interpretation |