Assessing regression-based deep learning for river ice estimation from drone images
Bibliographic record
Abstract
The concentration of frazil ice, crucial to the development of river ice covers and the numerical modeling of ice cover development, is challenging to measure in situ. Remote sensing using deep neural networks on images of frazil drift ice taken from drone is promising but faces challenges due to limited annotated datasets and difficulty in visually distinguishing ice types and boundaries. In this work, a method for acquiring and processing optical drone river ice images was developed to estimate the concentration of frazil drift ice, mostly in frazil slush form. Drone images were acquired on four mesoscale rivers (widths of ≈ 30 to 100 m) situated in the south of the Province of Quebec, Canada during the 2022–2023 and the 2023–2024 winter. A first Convolutional Neural Network was trained to perform an initial classification. This Convolutional Neural Network, the static ice model, was trained to segment the images in four classes: water, static ice, trees above water and other. Despite a few minor classification errors, the model was used to estimate the extent of static ice cover. Once the initial classification was made, the frazil drift ice concentration was estimated by taking into account only the flow zone. To do so, two Convolutional Neural Networks were trained with the same dataset but annotated with two different techniques: semantic segmentation and regression. Following the analysis of the results, it was concluded that regression is highly promising for estimating frazil drift ice concentration, particularly when the ice is in slush form and at high concentrations. The differences between the concentrations obtained using this method and those obtained manually are quite small (between 0 % and 2.2 %). With the same annotation effort as regression, the segmentation technique shows higher deviations (between 0.1 % and 9.4 %). The segmentation trained model encounters challenges in accurately identifying water areas surrounded by frazil and tend to extend frazil boundaries beyond their actual limits, which lead to an overestimation of the frazil drift ice concentration. These results confirm the potential of using drone imagery to train a regression-annotated Convolutional Neural Network for estimating frazil surface concentration in mesoscale rivers.
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 distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
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 teacher head, 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".