Automated Palynological Analysis System: Integrating Deep Metric Learning and $U^{2}$-Net Detection in $H\infty$ bright field microscopy

B. Mu\~noz; C. Toro; I. Lamas; I. Sanhueza; J. Staforelli-Vivanco; J. Troncoso; L. Viafora; M. Rondanelli-Reyes; P. Coelho; R. Jofr\'e

arxiv: 2604.16743 · v1 · submitted 2026-04-17 · 💻 cs.CV · physics.optics

Automated Palynological Analysis System: Integrating Deep Metric Learning and U²-Net Detection in Hinfty bright field microscopy

J. Staforelli-Vivanco , R. Jofr\'e , B. Mu\~noz , V. Salamanca , P. Coelho , I. Sanhueza , L. Viafora , C. Toro

show 3 more authors

J. Troncoso M. Rondanelli-Reyes I. Lamas

This is my paper

Pith reviewed 2026-05-10 08:07 UTC · model grok-4.3

classification 💻 cs.CV physics.optics

keywords automated palynologypollen classificationU2-Netdeep metric learningDINOv2bright field microscopymelissopalynologyimage detection

0 comments

The pith

An integrated deep learning system automates pollen counting and classification with 95.8% recall and sixfold speedup over manual analysis.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces an automated system for melissopalynology that uses bright-field microscopy with H∞ control to process honey samples from Chile's Bio Bio region. It applies U²-Net to detect pollen grains and trains a DINOv2 Vision Transformer with deep metric learning to classify them while providing interpretable attention maps. This addresses the time-consuming and subjective nature of traditional manual pollen analysis that takes 4-6 hours per sample. A sympathetic reader cares because faster, consistent analysis could support more widespread use in food quality control and ecological studies.

Core claim

The central discovery is that combining U²-Net for salient object detection with deep metric learning on a DINOv2 backbone, integrated with gradient-weighted attention, enables precise automated counting, classification, and morphological analysis of pollen grains, reaching 95.8% classification recall and a 6x processing speedup compared to manual expert analysis.

What carries the argument

The U²-Net model for detecting salient pollen objects paired with a DINOv2 Vision Transformer trained through deep metric learning for classification, augmented by Gradient-Weighted Attention to annotate diagnostic features.

Load-bearing premise

The models trained on Bio Bio region pollen images will maintain high accuracy when applied to new samples, different imaging conditions, or pollen from other geographic areas.

What would settle it

Collect a test set of pollen images from a different region or under altered microscope settings and measure if the classification recall falls significantly below 95.8%.

Figures

Figures reproduced from arXiv: 2604.16743 by B. Mu\~noz, C. Toro, I. Lamas, I. Sanhueza, J. Staforelli-Vivanco, J. Troncoso, L. Viafora, M. Rondanelli-Reyes, P. Coelho, R. Jofr\'e, V. Salamanca.

**Figure 3.** Figure 3: a. The automated system is orchestrated by a central processing workstation that executes the deep learning pipeline for image analysis. To manage high-throughput spatial scanning, it interfaces via a serial data bus with a microcontroller, which issues precise commands to actuate the XY motorized stage. Simultaneously, an intelligent autofocus mechanism dynamically compensates for Z-axis drift, ensu… view at source ↗

**Figure 4.** Figure 4: Inference preview examples (DINOv2 ViT-s). Centroid-based classification. By normalizing the gray background (128), the resulting pixel values are almost zero. This means that the background does not trigger the neural network, eliminating the need for computationally expensive masked pooling. 3.2 Image annotation, segmentation, and background To optimize storage and computational efficiency, the system … view at source ↗

**Figure 5.** Figure 5: Examples of four synthetic augmentation performed to left starting from the originals 1. Acaena splendens (bidibid), 2. Cucurbita pepo (squash), 3. Embothrium coccineum (Chilean firebush) and 4. Lithrea caustica (litre tree) 3.4 Embedding Model (AnalogyNet + DINOv2) To construct the robust embedding space required for pollen classification, our proposed AnalogyNet architecture replaces standard Convoluti… view at source ↗

**Figure 6.** Figure 6: Performance comparison. Several [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗

**Figure 7.** Figure 7: Model comparison based on Latent Space Analysis: 2D t-SNE S 128 per class (a, b, c and d) and density of 685 samples (e, f, g and h). Blocks apply for ConvNextV2-Tiny (a,e), DeiT-s (b,f), DINOv2 ViT-s (c,g) and ResNet-50 (d,h) [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗

**Figure 11.** Figure 11: Train and validation balance in DINOv2 ViT-s model [PITH_FULL_IMAGE:figures/full_fig_p009_11.png] view at source ↗

**Figure 9.** Figure 9: Hypersphere embedding representation in DINOv2 ViT-s. a. and b. : Two 3D PCA views. c. d. and e. Three 3D t-SNE angular perspectives . Block: Color code by class. Clear geometric separability validating the Deep Metric Learning hypothesis. This Figure explicitly illustrates the geometric constraint imposed at the end of the DINOv2 model pipeline. Instead of simply showing clusters floating in 3D space, it… view at source ↗

**Figure 12.** Figure 12: a. Global Confusion Matrix in DINOv2 ViT-s (95.1% Accuracy). b. Normalized confusion matrix. 9 [PITH_FULL_IMAGE:figures/full_fig_p009_12.png] view at source ↗

**Figure 13.** Figure 13: a. Distance distribution in DINOv2 ViT-s. Intra and inter-class distance. b. Intraclass distance by class. (powered by the U 2 -Net architecture), which localizes N individual pollen grains within the full field of view, extracting their respective bounding boxes, geometric contours, and saliency scores. Subsequently, the pipeline iterates over each detected grain to extract a localized crop with a 20… view at source ↗

**Figure 14.** Figure 14: Gradient-Weighted Attention in DINOv2 ViT-s for endemic pollen samples. a. Lithrea caustica and b. Quillaja saponaria. Original images, Grad-Cam heat map and predicted euclidean distance with numerical values in the 128- dimensional latent metric space (S 127). Because the embeddings are L2-normalized to a unit hypersphere, these distances are mathematically tied to cosine similarity, ensuring consiste… view at source ↗

**Figure 15.** Figure 15: Gradient-Weighted Attention summary [PITH_FULL_IMAGE:figures/full_fig_p011_15.png] view at source ↗

**Figure 16.** Figure 16: Bars represent the mean morphological diameter (µ) calculated from the U 2 -Net salient object detection masks , with error bars indicating the standard deviation (σ) to reflect intraclass biological variability. The sizes were derived using a spatial resolution of 0.15 µm/px, consistent with the 60x oil immersion bright-field microscopy setup. Especies are ordered by mean size, ranging from the larg… view at source ↗

read the original abstract

Traditional melissopalynology is a time-consuming and subjective process, often taking 4-6 hours per sample. We present an automated, high-throughput microscopy system that integrates $H\infty$ robust mechanical control with advanced deep learning pipelines for the precise counting, classification, and morphological analysis of pollen grains from Bio Bio region in south central territory in Chile. Our system employs $U^{2}$-Net for salient object detection and a DINOv2 Vision Transformer backbone trained via Deep Metric Learning for classification. By integrating Gradient-Weighted Attention, the model provides human-interpretable texture and diagnostic feature annotations. The system achieves a 95.8$\%$ classification recall and a 6x processing speedup compared to manual expert analysis.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper applies known deep learning components to automate pollen analysis but provides too few experimental details to back up its performance claims.

read the letter

The paper applies known deep learning components to automate pollen analysis but provides too few experimental details to back up its performance claims. It integrates U²-Net for detecting grains in the images and DINOv2 with metric learning for classifying them, all tied to a controlled microscopy setup. The authors also include attention visualizations to show what features the model uses. This is a reasonable engineering effort for a specific applied problem in melissopalynology. The strength is the end-to-end system for a practical task. Traditional counting takes hours, so any reliable automation could help labs processing many samples. The regional focus on Bio Bio pollen is fine for a starting point. The soft spots are bigger. The abstract gives the 95.8% recall and 6x speedup without dataset sizes, splits, baselines, or validation steps. More critically, there is no testing on data from other regions or under changed imaging conditions. Performance on held-out data from the same source does not show robustness to the shifts that would occur in broader use. The speedup claim is also tied to this single pipeline. This is for practitioners in palynology or food science who want tools to speed up their work. It is not for readers seeking new methods or broad scientific insights. The approach is sensible, but the evidence does not yet support the headline numbers strongly enough. I would not bring it to a reading group. I would not cite it. It should go to peer review so that the missing details can be addressed and the generalization can be checked.

Referee Report

3 major / 1 minor

Summary. The manuscript presents an automated palynological analysis system for pollen grains from the Bio Bio region in Chile. It combines H∞ robust mechanical control in bright-field microscopy with U²-Net for salient object detection and a DINOv2 Vision Transformer backbone trained via deep metric learning for classification, with Gradient-Weighted Attention for interpretability. The central claims are a 95.8% classification recall and a 6x processing speedup relative to manual expert analysis.

Significance. If the reported performance is supported by proper validation, the work could meaningfully advance high-throughput automation of melissopalynology, reducing the 4-6 hour manual analysis time and subjectivity. The integration of robust control, modern self-supervised vision backbones, metric learning, and attention-based interpretability is a coherent technical contribution with potential utility in ecology, apiculture, and environmental monitoring.

major comments (3)

[Methods] Methods section: No dataset size, class distribution, train/test split, or validation protocol (e.g., k-fold, held-out set) is described for the DINOv2 + deep metric learning classifier. Without these, the 95.8% recall cannot be evaluated for generalization versus overfitting.
[Results] Results section: The 6x speedup claim lacks any description of the timing protocol, hardware, number of samples, or direct comparison to a documented manual workflow, rendering the quantitative advantage unverifiable.
[Results] No external test set or domain-shift experiments are reported. All evaluation appears to use internal splits from the same Bio Bio imaging distribution; this directly undermines claims of robustness to new geographic regions, illumination changes, or microscope variations under H∞ control.

minor comments (1)

[Abstract] The abstract and introduction would benefit from a one-sentence definition or reference for H∞ control and U²-Net to aid readers outside the immediate subfield.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below and will incorporate revisions to improve clarity and verifiability of the results.

read point-by-point responses

Referee: [Methods] Methods section: No dataset size, class distribution, train/test split, or validation protocol (e.g., k-fold, held-out set) is described for the DINOv2 + deep metric learning classifier. Without these, the 95.8% recall cannot be evaluated for generalization versus overfitting.

Authors: We agree that these critical details are omitted from the current Methods section. In the revised manuscript we will add a dedicated subsection that reports the total number of images and samples collected from the Bio Bio region, the per-class distribution, the train/test split ratios (including any stratification), and the validation protocol (e.g., k-fold cross-validation or held-out test set) used to compute the 95.8% recall. This addition will allow readers to assess generalization versus overfitting. revision: yes
Referee: [Results] Results section: The 6x speedup claim lacks any description of the timing protocol, hardware, number of samples, or direct comparison to a documented manual workflow, rendering the quantitative advantage unverifiable.

Authors: We acknowledge that the timing protocol, hardware platform, number of samples, and explicit comparison to the manual workflow are not described. The revised Results section will include these details: the hardware used for automated processing, the exact measurement protocol, the number of samples timed, and a side-by-side description of the manual expert workflow (4–6 h per sample) to substantiate the 6x speedup claim. revision: yes
Referee: [Results] No external test set or domain-shift experiments are reported. All evaluation appears to use internal splits from the same Bio Bio imaging distribution; this directly undermines claims of robustness to new geographic regions, illumination changes, or microscope variations under H∞ control.

Authors: The current evaluation uses internal splits from the Bio Bio dataset acquired under the H∞-controlled bright-field setup; the manuscript does not claim robustness to arbitrary new geographic regions or microscope hardware. To address the concern we will add explicit language in the Discussion clarifying the intended scope (Bio Bio region, controlled imaging conditions) and will discuss potential domain-shift limitations and the role of H∞ control in mitigating illumination and mechanical variations within this setting. revision: yes

Circularity Check

0 steps flagged

No circularity detected; empirical ML results with no derivation chain

full rationale

The manuscript reports an empirical pipeline (U²-Net detection + DINOv2 + deep metric learning) evaluated on Bio Bio pollen images, claiming 95.8% recall and 6x speedup. No first-principles derivation, uniqueness theorem, or mathematical prediction is presented that could reduce to its own inputs by construction. Performance numbers are standard held-out test metrics from a single-region dataset; they are not fitted parameters renamed as predictions, nor do they rely on self-citation load-bearing steps. The work is self-contained as an applied engineering report rather than a theoretical derivation.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the empirical success of standard deep-learning models applied to an unspecified regional dataset; no new physical laws or mathematical derivations are introduced.

free parameters (1)

Deep learning hyperparameters and training choices
U²-Net and DINOv2 training involve numerous tunable parameters whose values are not reported.

axioms (1)

domain assumption Deep neural networks trained on the available images can reliably detect and classify pollen grains in H∞ bright-field microscopy from the Bio Bio region.
This assumption directly supports the reported 95.8% recall figure.

pith-pipeline@v0.9.0 · 5487 in / 1252 out tokens · 43602 ms · 2026-05-10T08:07:03.893354+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages · 1 internal anchor

[1]

& Coelho, P

Machuca, G., Staforelli, J., Rondanelli-Reyes, M., Garces, R., Contreras-Trigo, B., Tapia, J., ... & Coelho, P. (2022). Hyperspectral mi- croscopy technology to detect syrups adul- teration of endemic guindo santo and quil- lay honey using machine-learning tools. Foods, 11(23), 3868

work page 2022
[2]

& Coelho, P

Jofre, R., Tapia, J., Troncoso, J., Staforelli, J., Sanhueza, I., Jara, A., ... & Coelho, P. (2025). YOLOv8-based on-the-fly classifier system for pollen analysis of Guindo Santo (Eucryphia glutinosa) honey and assessment of its monoflorality. Journal of Agriculture and Food Research, 19, 101665

work page 2025
[3]

Sevillano, V., & Aznarte, J. L. (2020). Precise pollen classification with deep learning.PLOS ONE,15(4), e0229751

work page 2020
[4]

DINOv2: Learning Robust Visual Features without Supervision

Oquab, M., Darcet, T., Moutakanni, T., Vo, H., Szafraniec, M., Khalidov, V., ... & Bo- janowski, P. (2023). DINOv2: Learning ro- bust visual features without supervision.arXiv preprint arXiv:2304.07193. 12 Table 1: Morphological and Texture Metrics for Pollen Identification Pollen Class Size(µm)Area(10 3 px2)Circ.(C)I.C.D.(µ±σ) 1.Acaena s.12.67±1.30 45.2±...

work page internal anchor Pith review Pith/arXiv arXiv 2023
[5]

Chefer, H., Gur, S., & Wolf, L. (2021). Trans- former interpretability beyond attention vi- sualization. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(pp. 782-791)

work page 2021
[6]

& Houlsby, N

Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., ... & Houlsby, N. (2021). An image is worth 16x16 words: Transformers for image recognition at scale. InInternational Conference on Learning Representations

work page 2021
[7]

Touvron, H., Cord, M., & J´ egou, H. (2022). DeiT III: Revenge of the ViT. InEuropean Conference on Computer Vision(pp. 516- 533). Springer, Cham

work page 2022
[8]

R., & Jagersand, M

Qin, X., Zhang, Z., Huang, C., Dehghan, M., Zaiane, O. R., & Jagersand, M. (2020). U²- Net: Going deeper with nested U-structure for salient object detection.Pattern Recognition, 106, 107404

work page 2020
[9]

Caron, M., Touvron, H., Misra, I., J´ egou, H., Mairal, J., Bojanowski, P., & Joulin, A. (2021). Emerging properties in self-supervised vision transformers. InProceedings of the IEEE/CVF International Conference on Com- puter Vision(pp. 9650-9660)

work page 2021
[10]

Musgrave, K., Belongie, S., & Lim, S. N. (2020). A metric learning reality check. InEu- ropean Conference on Computer Vision(pp. 681-699). Springer, Cham

work page 2020
[11]

Roth, K., Milbich, T., Sinha, S., Gupta, P., Ommer, B., & Cohen, J. P. (2022). Non-isotropy regularization for proxy-based deep metric learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(pp. 7420-7430)

work page 2022
[12]

Kim, S., Kim, D., Cho, M., & Kwak, S. (2023). HIER: Metric learning beyond class-level sim- ilarities. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(pp. 11956-11966)

work page 2023
[13]

Wang, X., Han, X., Huang, W., Dong, D., & Scott, M. R. (2019). Multi-similarity loss with general pair weighting for deep metric learn- ing. InProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recog- nition(pp. 5022-5030)

work page 2019
[14]

R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., & Batra, D

Selvaraju, R. R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., & Batra, D. (2017). Grad-CAM: Visual explanations from deep networks via gradient-based localization. In Proceedings of the IEEE international confer- ence on computer vision(pp. 618-626)

work page 2017
[15]

M., Carva- jal, R., Caba˜ na-Brunod, M., & Otero, M

Poulsen-Silva, E., Gordillo-Fuenzalida, F., Vel´ asquez, P., Llancalahuen, F. M., Carva- jal, R., Caba˜ na-Brunod, M., & Otero, M. C. (2023). Antimicrobial, antioxidant, and anti- inflammatory properties of monofloral honeys from Chile. Antioxidants, 12(9), 1785

work page 2023
[16]

M., & Rondanelli- Reyes, M

Garc´ ıa, S., Troncoso, J. M., & Rondanelli- Reyes, M. (2020). Study of honey according to botanical origin and physicochemical parame- ters in the Biob´ ıo Region, Chile. Chilean jour- nal of agricultural research, 80(4), 675-685

work page 2020
[17]

Figueroa-Flores, C., & San-Martin, P. (2023). Deep learning for Chilean native flora classi- fication: a comparative analysis. Frontiers in Plant Science, 14, 1211490

work page 2023
[18]

Staforelli-Vivanco, J., Salamanca-Levi, V., Jofr´ e-Cerda, R., Rondanelli-Reyes, M., & Lamas, I. (2026). Three-Dimensional Volumet- ric Reconstruction of Native Chilean Pollen via Lens-Free Digital In-line Holographic Mi- croscopy. arXiv preprint arXiv:2601.14205

work page arXiv 2026
[19]

Louveaux, J., Maurizio, A., & Vorwohl, G. (1978). Methods of melissopalynology.Bee World,59(4), 139-157

work page 1978
[20]

E., & Krzywinski, K

Faegri, K., Kaland, P. E., & Krzywinski, K. (1989). Textbook of pollen analysis

work page 1989
[21]

J., Barnes, B

Sohn, Y. J., Barnes, B. M., Howard, L., Sil- ver, R. M., Attota, R., & Stocker, M. T. 13 (2006, March). K¨ ohler illumination for high- resolution optical metrology. In Metrology, In- spection, and Process Control for Microlithog- raphy XX (Vol. 6152, pp. 1236-1244). SPIE

work page 2006
[22]

https://www.madcitylabs.com/cfocus.html 14

work page

[1] [1]

& Coelho, P

Machuca, G., Staforelli, J., Rondanelli-Reyes, M., Garces, R., Contreras-Trigo, B., Tapia, J., ... & Coelho, P. (2022). Hyperspectral mi- croscopy technology to detect syrups adul- teration of endemic guindo santo and quil- lay honey using machine-learning tools. Foods, 11(23), 3868

work page 2022

[2] [2]

& Coelho, P

Jofre, R., Tapia, J., Troncoso, J., Staforelli, J., Sanhueza, I., Jara, A., ... & Coelho, P. (2025). YOLOv8-based on-the-fly classifier system for pollen analysis of Guindo Santo (Eucryphia glutinosa) honey and assessment of its monoflorality. Journal of Agriculture and Food Research, 19, 101665

work page 2025

[3] [3]

Sevillano, V., & Aznarte, J. L. (2020). Precise pollen classification with deep learning.PLOS ONE,15(4), e0229751

work page 2020

[4] [4]

DINOv2: Learning Robust Visual Features without Supervision

Oquab, M., Darcet, T., Moutakanni, T., Vo, H., Szafraniec, M., Khalidov, V., ... & Bo- janowski, P. (2023). DINOv2: Learning ro- bust visual features without supervision.arXiv preprint arXiv:2304.07193. 12 Table 1: Morphological and Texture Metrics for Pollen Identification Pollen Class Size(µm)Area(10 3 px2)Circ.(C)I.C.D.(µ±σ) 1.Acaena s.12.67±1.30 45.2±...

work page internal anchor Pith review Pith/arXiv arXiv 2023

[5] [5]

Chefer, H., Gur, S., & Wolf, L. (2021). Trans- former interpretability beyond attention vi- sualization. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(pp. 782-791)

work page 2021

[6] [6]

& Houlsby, N

Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., ... & Houlsby, N. (2021). An image is worth 16x16 words: Transformers for image recognition at scale. InInternational Conference on Learning Representations

work page 2021

[7] [7]

Touvron, H., Cord, M., & J´ egou, H. (2022). DeiT III: Revenge of the ViT. InEuropean Conference on Computer Vision(pp. 516- 533). Springer, Cham

work page 2022

[8] [8]

R., & Jagersand, M

Qin, X., Zhang, Z., Huang, C., Dehghan, M., Zaiane, O. R., & Jagersand, M. (2020). U²- Net: Going deeper with nested U-structure for salient object detection.Pattern Recognition, 106, 107404

work page 2020

[9] [9]

Caron, M., Touvron, H., Misra, I., J´ egou, H., Mairal, J., Bojanowski, P., & Joulin, A. (2021). Emerging properties in self-supervised vision transformers. InProceedings of the IEEE/CVF International Conference on Com- puter Vision(pp. 9650-9660)

work page 2021

[10] [10]

Musgrave, K., Belongie, S., & Lim, S. N. (2020). A metric learning reality check. InEu- ropean Conference on Computer Vision(pp. 681-699). Springer, Cham

work page 2020

[11] [11]

Roth, K., Milbich, T., Sinha, S., Gupta, P., Ommer, B., & Cohen, J. P. (2022). Non-isotropy regularization for proxy-based deep metric learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(pp. 7420-7430)

work page 2022

[12] [12]

Kim, S., Kim, D., Cho, M., & Kwak, S. (2023). HIER: Metric learning beyond class-level sim- ilarities. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(pp. 11956-11966)

work page 2023

[13] [13]

Wang, X., Han, X., Huang, W., Dong, D., & Scott, M. R. (2019). Multi-similarity loss with general pair weighting for deep metric learn- ing. InProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recog- nition(pp. 5022-5030)

work page 2019

[14] [14]

R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., & Batra, D

Selvaraju, R. R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., & Batra, D. (2017). Grad-CAM: Visual explanations from deep networks via gradient-based localization. In Proceedings of the IEEE international confer- ence on computer vision(pp. 618-626)

work page 2017

[15] [15]

M., Carva- jal, R., Caba˜ na-Brunod, M., & Otero, M

Poulsen-Silva, E., Gordillo-Fuenzalida, F., Vel´ asquez, P., Llancalahuen, F. M., Carva- jal, R., Caba˜ na-Brunod, M., & Otero, M. C. (2023). Antimicrobial, antioxidant, and anti- inflammatory properties of monofloral honeys from Chile. Antioxidants, 12(9), 1785

work page 2023

[16] [16]

M., & Rondanelli- Reyes, M

Garc´ ıa, S., Troncoso, J. M., & Rondanelli- Reyes, M. (2020). Study of honey according to botanical origin and physicochemical parame- ters in the Biob´ ıo Region, Chile. Chilean jour- nal of agricultural research, 80(4), 675-685

work page 2020

[17] [17]

Figueroa-Flores, C., & San-Martin, P. (2023). Deep learning for Chilean native flora classi- fication: a comparative analysis. Frontiers in Plant Science, 14, 1211490

work page 2023

[18] [18]

Staforelli-Vivanco, J., Salamanca-Levi, V., Jofr´ e-Cerda, R., Rondanelli-Reyes, M., & Lamas, I. (2026). Three-Dimensional Volumet- ric Reconstruction of Native Chilean Pollen via Lens-Free Digital In-line Holographic Mi- croscopy. arXiv preprint arXiv:2601.14205

work page arXiv 2026

[19] [19]

Louveaux, J., Maurizio, A., & Vorwohl, G. (1978). Methods of melissopalynology.Bee World,59(4), 139-157

work page 1978

[20] [20]

E., & Krzywinski, K

Faegri, K., Kaland, P. E., & Krzywinski, K. (1989). Textbook of pollen analysis

work page 1989

[21] [21]

J., Barnes, B

Sohn, Y. J., Barnes, B. M., Howard, L., Sil- ver, R. M., Attota, R., & Stocker, M. T. 13 (2006, March). K¨ ohler illumination for high- resolution optical metrology. In Metrology, In- spection, and Process Control for Microlithog- raphy XX (Vol. 6152, pp. 1236-1244). SPIE

work page 2006

[22] [22]

https://www.madcitylabs.com/cfocus.html 14

work page