Improved Accuracy of Stomata Micrograph Classification Using

YOLOv11 with Gamma Correction and CLAHE

Yusri Simang

, Indrabayu

and Siti Andini Utiarahman

Department of Informatics Engineering, Hasanuddin University, Makassar, Indonesia

Department of Electrical Engineering, Hasanuddin University, Makassar, Indonesia

Keywords: Micrograph, Stomata, YOLOv11, Gamma Correction, CLAHE.

Abstract: This study aims to improve the accuracy of stomata classification in micrograph images through the You Only

Look Once version 11 (YOLOv11) algorithm combined with Gamma Correction and Contrast Limited

Adaptive Histogram Equalization (CLAHE) as preprocessing techniques. Microscopic images often suffer

from uneven illumination and weak contrast, which hinder automatic stomata detection. Gamma Correction

adjusts image brightness non-linearly, while CLAHE enhances local contrast without amplifying noise.

Stomata images were captured using a binocular microscope, annotated based on morphology, and divided

into training, validation, and test sets for model development. Experimental results show that preprocessing

improves YOLOv11 performance, with precision increasing from 0.93 to 0.94, recall from 0.91 to 0.92, and

mean Average Precision (mAP) at 50% intersection over union from 0.95 to 0.96. For the Graminoid class,

precision increased from 0.84 to 0.86, recall from 0.88 to 0.89, and mAP@50 from 0.82 to 0.83. The Peak

Signal-to-Noise Ratio (PSNR) also improved from 27.9–28.35 dB to 28.44–29.40 dB, indicating better image

quality. These results demonstrate that the integration of Gamma Correction and CLAHE effectively enhances

image clarity and improves stomata detection performance, supporting more reliable and efficient automation

in botanical analysis.

1 INTRODUCTION

Stomata are microscopic pores on the surface of plant

leaves and stems that play a vital role in gas exchange

and the regulation of water loss through transpiration.

Their number, size, and morphology are important

parameters in plant physiology, ecological

adaptation, and taxonomic classification (Haworth et

al., 2023). Traditionally, stomata are identified

manually using light microscopy, a process that is

time-consuming, subjective, and highly dependent on

individual expertise. The challenge becomes greater

when micrograph images suffer from blurriness,

uneven illumination, or low contrast, which

significantly hampers accurate morphological

analysis (Gibbs & Burgess, 2024). To overcome these

limitations, computer vision and deep learning

approaches are increasingly applied for automatic

stomata classification.

You Only Look Once (YOLO) is a family of one-

stage object detection algorithms that perform

https://orcid.org/0000-0003-2026-1809

classification and localization simultaneously,

enabling real-time analysis. The latest version,

YOLOv11, introduces modules such as C2f, attention

mechanisms, and Spatial Pyramid Pooling Fast

(SPPF), which improve small-object detection and

computational efficiency. Several studies have

demonstrated its effectiveness for stomatal analysis

Zhang et al. (2021) combined YOLO with superpixel

segmentation for stomata recognition, Zhang et al.

(2023) proposed the DeepRSD method for randomly

oriented stomata, while Yang et al. (2025) developed

StomaYOLO for maize stomata detection. Beyond

stomata, YOLO has also been applied successfully in

plant disease classification (Alhwaiti et al., 2025).

These studies confirm YOLO’s robustness,

supporting its adoption in this research.

However, detection performance with YOLO

strongly depends on input image quality. Microscopic

images often suffer from uneven lighting, subtle

noise, and low contrast that obscure stomatal

structures. To address this, image enhancement

Simang, Y., Indrabayu, and Utiarahman, S. A.

Improved Accuracy of Stomata Micrograph Classiﬁcation Using YOLOv11 with Gamma Correction and CLAHE.

DOI: 10.5220/0014275100004928

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Innovations in Information and Engineering Technology (RITECH 2025), pages 68-75

ISBN: 978-989-758-784-9

methods were employed before YOLOv11 detection.

Gamma Correction was used to normalize

illumination by amplifying darker regions without

overexposing bright areas, as shown effective in low-

light enhancement (Jeon et al., 2024; Wang et al.,

2023; Mim et al., 2025). Meanwhile, Contrast

Limited Adaptive Histogram Equalization (CLAHE)

was applied to enhance local contrast while

controlling noise, a technique widely validated in

biomedical and plant imaging (Liu et al., 2021; Gibbs

et al., 2021; Narla et al., 2024; Buriboev et al., 2024).

The integration of Gamma Correction and

CLAHE has also proven beneficial. For example,

Chang et al. (2018) reported that dual gamma

correction with CLAHE improved visual quality in

low-light images, while Benchabane & Charif (2025)

achieved a 6–9% accuracy gain in COVID-19 X-ray

classification using this combination. Mim et al.

(2025) further confirmed the effectiveness of

integrating these techniques for structural clarity and

classification performance. Despite these advances,

no studies have explicitly incorporated Gamma

Correction and CLAHE into the YOLOv11 pipeline

for stomatal classification. This research therefore

proposes an automatic stomata classification system

based on YOLOv11 with integrated Gamma

Correction and CLAHE preprocessing, aimed at

improving accuracy in tropical leaf micrographs and

supporting the automation of botanical analysis and

precision agriculture.

2 METHODOLOGY

This study focuses on the development of a stomata

classification and detection system using a deep

learning algorithm under uneven lighting conditions

in stomata micrograph images. The dataset used in

this study consisted of 312 micrograph images of

stomata from herbal plant leaves, captured using a

biological binocular microscope. These images were

distributed across four stomata classes: Anomocytic

(76 images), Diacytic (84 images), Graminoid (80

images), and Paracytic (72 images). The dataset was

divided into training (218 images, 70%), validation

(62 images, 20%), and testing (31 images, 10%).

To enhance the training set and reduce overfitting,

data augmentation was applied. For each training

image, three additional augmented images were

generated by applying 90° clockwise rotation, 90°

counter-clockwise rotation, and a flipped orientation.

This process expanded the training set from 218 to

588 images. By introducing these geometric

variations, the model was exposed to stomata with

different orientations, ensuring better class balance

and improving robustness in recognizing

morphological structures under diverse conditions.

Following the annotation process, bounding boxes

and class labels were assigned to each stomata

instance. Since a single micrograph may contain

multiple stomata, the number of annotated instances

exceeds the total number of images. Table 1

summarizes the distribution of labels across the four

stomata classes.

After dataset preparation and annotation, the

workflow of the proposed system, including

preprocessing, model training, and evaluation, is

illustrated in Figure 1.

Figure 1: System Design Flow.

2.1 Training Phase

During the training phase, the YOLOv11 model was

fine-tuned using the Ultralytics framework. The pre-

processed dataset (Gamma Correction and CLAHE

applied) consisted of 588 augmented training images,

62 validation images, and 31 test images. Training

was performed for 100 epochs with a batch size of 4

and an input resolution of 640 × 640 pixels. The

learning rate was initialized at 0.001 with the default

Improved Accuracy of Stomata Micrograph Classiﬁcation Using YOLOv11 with Gamma Correction and CLAHE

optimizer provided by Ultralytics. All experiments

were executed on a workstation equipped with an

NVIDIA RTX 4070 GPU (8 GB VRAM). The model

checkpoint with the highest validation mAP@50 was

saved for subsequent evaluation.

2.1.1 Data Acquisition

Stomata images were acquired using a biological

binocular microscope at 3× and 4× magnification,

covering the four stomata types analyzed in this

study. The samples were obtained from four different

herbal plants, with one representative species used for

each stomata type. The process of data acquisition as

shown in Figure 2.

Figure 2: Stomata micrograph data acquisition.

2.1.2 Preprocessing

The image was pre-processed using Gamma

Correction (γ = 0.9) to non-linearly adjust brightness,

making fine stomatal structures more visible in darker

regions. In addition, CLAHE was applied with a clip

limit of 3.0 and a tile grid size of 8×8 to enhance local

contrast while preventing excessive noise. These

enhancements ensured that the input images

preserved morphological details critical for stomata

recognition. Figure 3 illustrates one of preprocessing

result.

Figure 3: Preprocessing results.

2.1.3 Annotations

Stomata are given bounding boxes and morphological

labels to train the YOLOv11 model to recognize and

distinguish stomata types.

2.1.4 Split Data

The dataset was split into training, validation, and

testing subsets for model training and evaluation as

shown in Table 1.

Table 1: Distribution of stomata labels.

Class

Labels

Train

Labels

Validation

Labels

Testin

Label

count

Anomocyti

3679 339 230 4248

Diacytic 4925 770 141 5836

Graminoid 3532 520 217 4269

Paracytic 4618 266 355 5239

Total 16754 1895 943 19592

2.1.5 Data Augmentation

In the Figure 4, each original stomata image was

augmented into three variations: one rotated 90°

clockwise, one 90° counterclockwise, and one kept

unchanged. This increased the dataset from 219 to

590 images, helping the YOLOv11 model generalize

better by simulating different microscope

orientations.

Figure 4: Data augmentation results.

RITECH 2025 - The International Conference on Research and Innovations in Information and Engineering Technology

2.1.6 Training Model YOLOv11

The pre-processed and annotated images were then

used to train YOLOv11. This integration allowed the

model to learn more discriminative visual features

from images with improved lighting and contrast,

leading to better stomata classification performance.

2.1.7 Build Model

The model is stored after validation and is ready to be

used for classification and detection of stomata in test

data.

2.2 Testing Phase

In the testing phase, the trained YOLOv11 model was

evaluated on the held-out test set of 31 images, which

had been preprocessed with Gamma Correction and

CLAHE. No augmentation or annotation was applied

during testing to ensure fairness. The model

predictions were compared with ground-truth labels

using precision, recall, mAP@50, mAP@50–95, and

F1-score as evaluation metrics.

2.2.1 Input Data

The images tested are images that have never been

used before and are entered into the system to

evaluate the final performance of the model. The

results of this stage provide an overview of the extent

to which the model is capable of working on

completely new data.

2.2.2 Preprocessing (Testing)

In the testing phase, only Gamma Correction and

CLAHE were applied to the images. Unlike the

training phase, no annotation or augmentation was

performed, as the goal was to evaluate the model on

unseen images with enhanced visibility and contrast,

ensuring a fair and unbiased evaluation.

2.2.3 Yolov11 Stomata Detection &

Classification

The trained YOLOv11 model directly received the

pre-processed test images and performed stomata

detection and classification. By using enhanced

images, the model was able to achieve higher

precision and recall in identifying small stomatal

structures.

2.2.4 Evaluation

To assess the performance of the proposed YOLOv11

model, detection results were compared against the

ground truth annotations using widely adopted

evaluation metrics, including Accuracy, Precision,

Recall, mean Average Precision (mAP), and Peak

Signal-to-Noise Ratio (PSNR).

To evaluate the performance of the YOLOv11

model in detecting various types of stomata under

different image processing scenarios (without

preprocessing, with Gamma Correction, and with

CLAHE), a confusion matrix was employed as the

primary evaluation technique (Figure 5). The

confusion matrix provides a detailed summary of

classification outcomes by categorizing predictions

into four main groups: True Positive (TP),

representing stomata objects that were correctly

detected and classified by the model; False Negative

(FN), referring to stomata objects that were missed

and not detected by the model; False Positive (FP),

representing non-stomata regions that were

incorrectly classified as stomata; and True Negative

(TN), indicating non-stomata regions that were

correctly identified as not belonging to the stomata

class.

Figure 5: Confusion Matrix.

Where TP (True Positive) denotes the number of

correctly detected stomata, TN (True Negative)

denotes the number of correctly rejected non-stomata

regions, FP (False Positive) represents the number of

incorrectly detected regions, and FN (False Negative)

represents the number of missed stomata.

Precision (1) measures the proportion of

correctly predicted stomata among all detected

regions.

𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 =

𝑇𝑃

𝐹𝑃

)

Improved Accuracy of Stomata Micrograph Classiﬁcation Using YOLOv11 with Gamma Correction and CLAHE

Recall (2) measures the proportion of correctly

detected stomata with respect to all actual stomata in

the ground truth.

𝑅𝑒𝑐𝑎𝑙𝑙 =

𝑇𝑃

𝐹𝑁

(2)

Where AP

represents the Average Precision for

class i, and N is the total number of classes. The mAP

(3) is widely used in object detection to evaluate the

overall detection performance across all classes.

𝑚𝐴𝑃 =

𝑁

𝐴𝑃







(3)

Where MAX

is the maximum possible pixel

value (255 for 8-bit images), and MSE is the Mean

Squared Error between the original image and the

processed image. PSNR (4) quantifies the quality of

image enhancement during preprocessing and

detection.

𝑃𝑆𝑁𝑅 = 10 ∙ 𝑙𝑜𝑔



(

𝑀𝐴𝑋





𝑀

𝑆

𝐸

)

(4)

3 RESULTS AND DISCUSSION

3.1 Peak Signal-to-Noise Ratio (PSNR)

Evaluation

The PSNR values in all three images showed an

increase after preprocessing using Gamma Correction

and CLAHE. In Table 2, the initial PSNR was in the

range of 27.9–28.35 dB and increased to 28.44–29.40

dB after preprocessing. This increase indicates an

improvement in the visual quality of the image,

especially in terms of lighting and local contrast,

without compromising important details. This proves

that the preprocessing technique used is effective in

clarifying the structure of the stomata, thus

supporting the automatic detection process by the

YOLOv11 model more accurately.

Table 2: PSNR evaluation.

Gamma

PSNR

(dB)

Gamma +

CLAHE

PSNR

(dB)

27.96

28.33

28.01

28.69

27.88

28.63

28.43

29.32

3.2 YOLOv11 Training Results with

Gamma Correction + CLAHE

The training was conducted on an NVIDIA RTX

4070 GPU (8 GB VRAM) using the Ultralytics

YOLO framework. Each epoch required ~2.5

minutes, with a total training time of 4.2 hours for 100

epochs, and inference reached 45 FPS (≈22 ms per

image), showing near real-time suitability. As shown

in Table 3, preprocessing improved YOLOv11

performance compared to the baseline (P = 0.93, R =

0.91, mAP@50 = 0.95, mAP@50–95 = 0.54).

Gamma Correction slightly raised recall and

mAP@50–95, CLAHE increased precision, and their

combination gave the best results (P = 0.94, R = 0.92,

mAP@50 = 0.96, mAP@50–95 = 0.55). Accuracy

was calculated as.

𝐴

𝑐𝑐𝑢𝑟𝑎𝑐𝑦 =

𝑇𝑃 + 𝑇𝑁

𝑇𝑃+𝑇𝑁+𝐹𝑃+𝐹𝑁

)

The proposed method achieved 95% accuracy,

higher than Zhang et al. (2021) with 91.6%, Gibbs et

al. (2021) with 90.2%, Zhang et al. (2023) with

RITECH 2025 - The International Conference on Research and Innovations in Information and Engineering Technology

94.3%, and Utiarahman (2025) with 93.5%. These

results confirm that Gamma + CLAHE preprocessing

enhances YOLOv11 as shown in Figure 7, providing

the better balance of precision, recall, and accuracy

rather than without Gamma + CLAHE (Figure 6)..

Table 3: Results of YOLOv11 Original and YOLOv11 Training with Gamma + CLAHE.

Study / Method Dataset / Target Approach Reported Performance

Zhang et al., 2021

Stomata micrographs

(various plants)

YOLO + entropy rate

superpixel

segmentation

Accuracy = 93%

Zhang et al., 2023 Corn leaves DeepRSD Accuracy= 94.3%

Gibbs et al., 2021

Wheat stomata

micrographs

CNN-based Accuracy > 90%

Utiarahman, 2025 Herbal plants

Benchmark

YOLOv8–v11

YOLOv11: Precision = 0.933, Recall =

0.886, mAP@50 = 0.951, mAP@50–95 =

0.515

This study (YOLOv11

aseline)

Herbal plant stomata (4

types)

YOLOv11 without

reprocessing

Accuracy = 94% Precision = 0.93, Recall =

0.91, mAP@50 = 0.95, mAP@50

–

95 = 0.54

This study (YOLOv11 +

Gamma

)

Herbal plant stomata (4

)

YOLOv11 with

Gamma

Accuracy = 95% Precision = 0.93, Recall =

0.92, mAP

50 = 0.95, mAP

–

95 = 0.55

This study (YOLOv11 +

CLAHE

)

Herbal plant stomata (4

)

YOLOv11 with

CLAHE

Accuracy = 94% Precision = 0.94, Recall =

0.91, mAP

50 = 0.94, mAP

–

95 = 0.54

This study (YOLOv11 +

Gamma + CLAHE)

Herbal plant stomata (4

types)

YOLOv11 with

Gamma Correction +

CLAHE

Accuracy = 95% Precision = 0.94, Recall =

0.92, mAP@50 = 0.96, mAP@50–95 = 0.55

Figure 6: Graph of the results of the original YOLOv11 training (mAP 50, mAP 50-95, precision, and recall).

Figure 7: Graph of YOLOv11 training results with Gamma Correction + CLAHE (mAP 50, mAP 50-95, precision, and

recall).

Improved Accuracy of Stomata Micrograph Classiﬁcation Using YOLOv11 with Gamma Correction and CLAHE

3.3 YOLOv11 Testing Results with

Gamma + CLAHE

The test results in Table 4 showed that the YOLOv11

model with Gamma Correction and CLAHE

preprocessing outperformed the original model. In the

Anomocytic and Paracytic (AD) classes, there was an

increase in precision, mAP, and accuracy. For the

Graminoid class, recall increased from 0.88 to 0.89

and precision from 0.84 to 0.86, indicating improved

detection. Meanwhile, the Diacytic class achieved

consistently high results on both models (precision

0.95–0.96; accuracy 0.99), indicating optimal

detection even without preprocessing.

Table 4: YOLOv11 Testing Results with Gamma + CLAHE.

Metho

Class TP FP FN TN Precision Recall Accurac

mAP 50 mAP 50-95

YOLOv11

Original

Anomocytic 202 28 32 787 0.91 0.83 0.94 0.93 0.52

Diac

tic 138 3 7 901 0.95 0.97 0.99 0.98 0.57

Graminoi

200 17 44 788 0.84 0.88 0.93 0.82 0.43

Parac

tic 352 3 23 671 0.95 0.98 0.98 0.99 0.59

YOLOv11 +

Gamma

Correction

Anomocytic 201 29 30 797 0.90 0.83 0.94 0.93 0.53

Diacytic 138 3 9 907 0.95 0.97 0.99 0.99 0.58

Graminoi

200 17 50 790 0.85 0.90 0.94 0.83 0.44

Parac

tic 350 5 26 676 0.95 0.98 0.97 0.99 0.60

YOLOv11 +

CLAHE

Anomoc

tic 199 31 33 795 0.92 0.83 0.94 0.92 0.52

Diacytic 138 18 47 794 0.96 0.98 0.99 0.99 0.57

Graminoi

199 18 47 794 0.86 0.87 0.94 0.82 0.43

Paracytic 351 4 27 676 0.95 0.98 0.97 0.99 0.59

YOLOv11 +

Gamma

Correction +

CLAHE

Anomoc

tic 204 26 32 793 0.91 0.85 0.95 0.94 0.52

Diac

tic 138 3 10 904 0.96 0.98 0.99 0.99 0.58

Graminoi

201 16 10 790 0.86 0.89 0.94 0.83 0.43

Paracytic 351 4 22 678 0.95 0.98 0.98 0.99 0.60

3.4 Results of Gamma+CLAHE

Application

Figure 8 presents the results of stomata detection on

micrograph images processed with Gamma

Correction and CLAHE. The YOLOv11 model

successfully detects stomata, as indicated by the

bounding boxes labeled "diacytic" and "paracytic" in

the relevant regions. CLAHE enhances local contrast,

making stomatal structures clearer even in previously

blurry or dim areas, while Gamma Correction

balances overall illumination across the image.

Figure 8: Comparison of the application of Gamma

Correction + CLAHE.

4 CONCLUSIONS

This study developed a YOLOv11-based automatic

stomata classification system with Gamma

Correction and CLAHE preprocessing. The results

showed improved image quality (higher PSNR) and

better detection performance in terms of precision,

recall, and mAP. The model provided more stable and

accurate detection of different stomata types,

confirming the benefit of combining image

enhancement with deep learning for botanical

analysis.

However, the study has limitations, including a

relatively small dataset restricted to four stomata

types and moderate performance gains from

preprocessing. Although YOLOv11 achieved

acceptable inference speed, further optimization is

needed for large-scale or real-time use.

Future work should involve real-time

experiments, the integration of attention mechanisms

to enhance feature extraction, and expansion of the

dataset with additional plant species to improve

robustness and generalizability. These efforts are

expected to increase both the efficiency and accuracy

of automatic stomata detection for broader

applications in plant science and precision

agriculture.

RITECH 2025 - The International Conference on Research and Innovations in Information and Engineering Technology

ACKNOWLEDGEMENTS

The authors express their gratitude to colleagues and

collaborators for their valuable support, and to the

laboratory staff for assistance during data collection.

Special thanks are also due to the reviewers for their

constructive feedback, which improved the quality of

this manuscript. The use of generative AI tools is

acknowledged for enhancing the language and

readability of this paper under human oversight.

REFERENCES

Alhwaiti, Y., Khan, M., Asim, M., Siddiqi, M. H., Ishaq,

M., & Alruwaili, M. (2025). Leveraging YOLO deep

learning models to enhance plant disease identification.

Scientific Reports, 15(1).

https://doi.org/10.1038/s41598-025-92143-0

Benchabane, A., & Charif, F. (2025). Enhanced COVID-19

Detection Through Combined Image Enhancement and

Deep Learning Techniques. Informatica (Slovenia),

49(16), 67–76.

https://doi.org/10.31449/inf.v49i16.5869

Buriboev, A. S., Khashimov, A., Abduvaitov, A., & Jeon,

H. S. (2024). CNN-Based Kidney Segmentation Using

a Modified CLAHE Algorithm. Sensors, 24(23).

https://doi.org/10.3390/s24237703

Gibbs, J. A., & Burgess, A. J. (2024). Application of deep

learning for the analysis of stomata: a review of current

methods and future directions. In Journal of

Experimental Botany (Vol. 75, Issue 21, pp. 6704–

6718). Oxford University Press.

https://doi.org/10.1093/jxb/erae207

Gibbs, J. A., Mcausland, L., Robles-Zazueta, C. A.,

Murchie, E. H., & Burgess, A. J. (2021). A Deep

Learning Method for Fully Automatic Stomatal

Morphometry and Maximal Conductance Estimation.

Frontiers in Plant Science, 12.

https://doi.org/10.3389/fpls.2021.780180

Haworth, M., Marino, G., Materassi, A., Raschi, A., Scutt,

C. P., & Centritto, M. (2023). The functional

significance of the stomatal size to density relationship:

Interaction with atmospheric [CO2] and role in plant

physiological behaviour. In Science of the Total

Environment (Vol. 863). Elsevier B.V.

https://doi.org/10.1016/j.scitotenv.2022.160908

Jeon, J. J., Park, J. Y., & Eom, I. K. (2024). Low-light

image enhancement using gamma correction prior in

mixed color spaces. Pattern Recognition, 146.

https://doi.org/10.1016/j.patcog.2023.110001

Liu, H., Zhao, Z., & She, Q. (2021). Self-supervised ECG

pre-training. Biomedical Signal Processing and

Control, 70.

https://doi.org/10.1016/j.bspc.2021.103010

Mim, S. A., Cheng, J., & Zhang, J. (2025). Fusion-based

framework for low-light image enhancement using

BM3D filter and adaptive gamma correction. Optics

and Laser Technology, 192.

https://doi.org/10.1016/j.optlastec.2025.113662

Narla, V. L., Suresh, G., Rao, C. S., Awadh, M. Al, &

Hasan, N. (2024). A multimodal approach with firefly

based CLAHE and multiscale fusion for enhancing

underwater images. Scientific Reports, 14(1).

https://doi.org/10.1038/s41598-024-76468-w

Utiarahman, S. A., Indrabayu, & Nurtanio, I. (2025).

Performance Evaluation of YOLOv8 to YOLOv11 for

Accurate Detection and Classification of Stomata in

Microscopic Images of Herbal Plants. Proceedings of

the NMITCON 2025 International Conference. IEEE.

https://doi.org/10.1109/NMITCON65824.2025.11188

393

Wang, Y., Liu, Z., Liu, J., Xu, S., & Liu, S. (2023). Low-

Light Image Enhancement with Illumination-Aware

Gamma Correction and Complete Image Modelling

Network. http://arxiv.org/abs/2308.08220

Yang, Z., Liao, Y., Chen, Z., Lin, Z., Huang, W., Liu, Y.,

Liu, Y., Fan, Y., Xu, J., Xu, L., & Mu, J. (2025).

StomaYOLO: A Lightweight Maize Phenotypic

Stomatal Cell Detector Based on Multi-Task Training.

Plants, 14(13). https://doi.org/10.3390/plants14132070

Zhang, F., Ren, F., & Li, J. (2021). Automatic stomata

recognition and measurement based on improved

YOLO deep learning model and entropy rate superpixel

algorithm. Ecological Informatics, 68, 101521.

https://doi.org/10.1016/j.ecoinf.2021.101521

Zhang, F., Wang, B., Lu, F., & Zhang, X. (2023). Rotating

Stomata Measurement Based on Anchor-Free Object

Detection and Stomata Conductance Calculation. Plant

Phenomics, 5.

https://doi.org/10.34133/plantphenomics.0106

Improved Accuracy of Stomata Micrograph Classiﬁcation Using YOLOv11 with Gamma Correction and CLAHE