Coconut Palm Tree Counting on Drone Images with Deep Object

Detection and Synthetic Training Data

Tobias Rohe

1 a

, Barbara B

ohm

, Michael K

olle

1 b

, Jonas Stein

1,2 c

, Robert M

uller

and Claudia Linnhoff-Popien

1 d

Mobile and Distributed Systems Group, LMU Munich, Germany

Aqarios GmbH, Germany

ﬁ

Keywords:

Object Detection, Synthetic Training Data, YOLOv7.

Abstract:

Drones have revolutionized various domains, including agriculture. Recent advances in deep learning have

propelled among other things object detection in computer vision. This study utilized YOLO, a real-time ob-

ject detector, to identify and count coconut palm trees in Ghanaian farm drone footage. The farm presented

has lost track of its trees due to different planting phases. While manual counting would be very tedious

and error-prone, accurately determining the number of trees is crucial for efﬁcient planning and management

of agricultural processes, especially for optimizing yields and predicting production. We assessed YOLO

for palm detection within a semi-automated framework, evaluated accuracy augmentations, and pondered its

potential for farmers. Data was captured in September 2022 via drones. To optimize YOLO with scarce

data, synthetic images were created for model training and validation. The YOLOv7 model, pretrained on

the COCO dataset (excluding coconut palms), was adapted using tailored data. Trees from footage were

repositioned on synthetic images, with testing on distinct authentic images. In our experiments, we adjusted

hyperparameters, improving YOLO’s mean average precision (mAP). We also tested various altitudes to de-

termine the best drone height. From an initial mAP@.5 of 0.65, we achieved 0.88, highlighting the value of

synthetic images in agricultural scenarios.

1 INTRODUCTION

Coconut farming, pivotal to West African economies,

offers both a sustainable livelihood and essential con-

tributions to regional food systems. However, chal-

lenges in monitoring the growth and count of co-

conut palm trees, increased by varied planting phases

and environmental conditions, pose signiﬁcant opera-

tional hindrances for larger farms. This paper presents

an innovative approach to address these challenges by

leveraging deep learning to detect and count coconut

palm trees using drone imagery, speciﬁcally focusing

on a farming project in the eastern region of Ghana.

Initiated in August 2021, the farming project

aimed to cultivate approximately 2,500 coconut palm

trees alongside other crops. The overarching goal was

not only to foster a sustainable family-run coconut

https://orcid.org/0009-0003-3283-0586

https://orcid.org/0000-0002-8472-9944

https://orcid.org/0000-0001-5727-9151

https://orcid.org/0000-0001-6284-9286

farming business but also to bolster employment op-

portunities and social beneﬁts for local communi-

ties, thereby intertwining traditional farming practices

with modern techniques. However, as the farm ex-

panded, maintaining an accurate count of the trees

became a daunting task, with manual surveys proving

both time-consuming and error-prone.

Addressing this, we explored the application of

computer vision, a subﬁeld of computer science fo-

cusing on replicating human vision system capa-

bilities, to detect and enumerate the coconut palm

trees. While classical object detection methods re-

lied on handcrafted features, recent advancements in

machine learning and deep learning have revolution-

ized this space, with techniques such as YOLO (You

Only Look Once) offering enhanced accuracy and ef-

ﬁciency.

This study delves into the application of the

YOLOv7 framework, released in 2022, to our spe-

ciﬁc use case. Beyond mere counting, future appli-

cations of this methodology might extend to discern-

ing the health of plants, thereby offering comprehen-

662

Rohe, T., Böhm, B., Kölle, M., Stein, J., Müller, R. and Linnhoff-Popien, C.

Coconut Palm Tree Counting on Drone Images with Deep Object Detection and Synthetic Training Data.

DOI: 10.5220/0013172900003890

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

In Proceedings of the 17th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2025) - Volume 3, pages 662-669

ISBN: 978-989-758-737-5; ISSN: 2184-433X

sive farm management solutions. By focusing on a

real-world problem, this work aims to bridge the gap

between advanced technical solutions and practical

agricultural challenges, setting the stage for more in-

tegrative, technology-driven farming practices in the

future. In summary, our contributions are:

1. We introduce a novel real world application: Fine-

tuning a deep object detector to count coconut

palm trees.

2. We show that performance can substantially be

increased by not only considering coconut palm

trees during training but also other plants. This

allows the object detector to better differentiate

between the plants it sees.

3. We show that it sufﬁces to train on synthetically

generated images and thereby eliminate the need

to manually label the images.

2 BACKGROUND

This section delves into the foundational concepts and

the current state of object detection. We start with

Convolutional neural networks. Subsequently, our fo-

cus shifts to object detection, a critical component to

address our use case. In this context, we enumer-

ate the four scenarios encountered during object de-

tection, especially when positioning bounding boxes.

We also elucidate the concept of Intersection-Over-

Union. Building on this, we discuss key performance

metrics in object detection, namely precision, recall,

and mean average precision. Concluding this section,

we present the YOLO architecture, emphasizing the

advancements in YOLOv7—a state-of-the-art object

detector. The experiments we discuss in Section 4

predominantly leverage this technology.

2.1 Convolutional Neural Networks

Convolutional Neural Networks (CNNs) have

emerged as a pivotal technology in the domain of

computer vision. Originating from the larger family

of Deep Neural Networks (DNNs), CNNs are specif-

ically tailored for image data, making them adept at

tasks such as face recognition, image classiﬁcation,

and object detection.

A standard CNN architecture comprises several

layers. The inaugural layer, the convolutional layer,

is pivotal for feature extraction. By utilizing ﬁlter ma-

trices, or kernels, this layer captures intricate patterns

such as colours and edges from an image. The unique

property of these ﬁlters is their translational invari-

ance, allowing objects to be recognized irrespective

of their spatial positioning in an image.

Following the convolutional layer is the pooling

layer, designed for dimensionality reduction. This di-

mension reduction not only decreases the computa-

tional burden but also helps in extracting dominant

features. Two common pooling methods exist: max

pooling and average pooling, which respectively cap-

ture the maximum and average values from a desig-

nated window in the input.

In the deeper sections of the network, fully-

connected layers serve the crucial role of integrating

features from previous layers and mapping them to

the desired output. These layers essentially form the

decision-making component of the CNN.

Training a CNN involves deﬁning its architec-

ture and then optimising it over several iterations,

known as epochs. Throughout this training phase,

the model’s internal parameters get reﬁned to enhance

prediction accuracy. This training is supervised, re-

quiring labelled datasets to guide the iterative minimi-

sation of prediction errors. Within the realm of CNNs,

renowned architectures include AlexNet, GoogLeNet,

and VGGNet.

Critical challenges in training CNNs encompass

phenomena like underﬁtting and overﬁtting. Address-

ing these challenges, often by tuning hyperparame-

ters, ensures that the trained model is both robust and

accurate in its predictions (Sevarac, 2021).

2.2 Object Detection

Computer vision, pivotal in numerous applications,

encompasses tasks such as image classiﬁcation, seg-

mentation, and object detection. Object detection

marries object recognition — identifying and clas-

sifying objects within media — with object local-

ization, which encapsulates these identiﬁed objects

within bounding boxes (Khandelwal, 2020). A key

metric here is the Intersection-over-Union (IoU),

which quantiﬁes the overlap between the predicted

bounding box and the ground truth — the actual an-

notated bounding box. It calculates the ratio of their

intersection to their union. IoU values range between

0 and 1: values close to 0 imply minimal overlap and

those nearing 1 signify accurate predictions (Anwer,

2022).

2.3 YOLO: You Only Look Once

This section delves into modern object detectors, fo-

cusing on YOLO (You Only Look Once). The ﬁeld

of object detection began with Region-based Convo-

lutional Network (R-CNN) in 2014. It proposed re-

Coconut Palm Tree Counting on Drone Images with Deep Object Detection and Synthetic Training Data

663

gions and fed them into a classiﬁer, a methodology

advanced by its successors: Fast R-CNN, Faster R-

CNN, and Mask R-CNN. These methods identiﬁed

potential regions of interest and subsequently classi-

ﬁed them for detection (Szeliski, 2022).

YOLO, introduced in 2016, diverges from this

two-step approach. It performs real-time object de-

tection by predicting bounding boxes and class prob-

abilities directly from images in a single evaluation,

enhancing speed and requiring fewer resources (Red-

mon et al., 2016). The model encompasses 24 convo-

lutional layers followed by 2 fully connected layers,

leveraging a rectiﬁed linear activation function in all

but the ﬁnal layer (Redmon et al., 2016). YOLO’s ar-

chitecture breaks the input image into an S × S grid,

making predictions encoded as an S × S × (B ∗ 5 +C)

tensor (Redmon et al., 2016).

Our experiments in section 4 employ YOLOv7,

released in 2022. This variant, pretrained on the MS

Coco Dataset with 80 object categories, lacks a cat-

egory for coconut palm trees, necessitating custom

data for our use case: counting coconut palm trees

in drone images. YOLOv7 boasts innovations like

the extended efﬁcient layer aggregation network (E-

ELAN) and a trainable bag of freebies, enhancing

speed and accuracy without elevating costs (Kukil and

Rath, 2022). The architecture, with around 37 mil-

lion parameters, features a backbone established in

the ﬁrst 50 layers.

3 METHOD

In this section our concept and the derived workﬂow

is described. Both are geared towards answering the

overarching question of the use case. The concept and

workﬂow was chosen as follows and reﬂects the typi-

cal phases of a deep learning project.

3.1 Data Acquisition

In September 2022, aerial drone images were cap-

tured in Ghana at heights of 10m, 25m, 45m, 70m,

and 85m. These 73 images, captured using a DJI

Mini 2 drone camera, have a resolution of 4000x2250

pixels. Given the common data scarcity challenge in

deep learning and the guideline suggesting roughly

5000 labeled examples per category for adequate per-

formance (Goodfellow I., 2016), we generated addi-

tional data synthetically. We based this on 13 se-

lected drone images, keeping the remaining images

for model testing.

3.2 Data Preparation and Generation

Data preparation follows a cross-validation approach,

partitioning the data into training, validation, and test

sets (Prince, 2023). Due to data limitations, we gener-

ated synthetic images to train YOLOv7. Using GIMP,

speciﬁc plants were extracted from the raw images,

while Dall-E produced the backgrounds (BG). A cus-

tom Python generator then assembled these compo-

nents into synthetic images by:

• Randomly selecting BGs and plants

• Applying random plant rotation, size, and ﬂip

• Positioning plants on the BG without overlaps

• Adjusting the plant count per BG based on con-

ﬁguration

This generator also produces YOLO-formatted la-

bel ﬁles, essential for training and validation. We up-

loaded the synthetic and test images to Google Drive,

organizing them into speciﬁc directories for compati-

bility with YOLOv7.

3.3 Metric

The precision-recall curve graphically illustrates the

trade-off between the two metrics for different con-

ﬁdence thresholds. As a reminder these metrics are

utilised to evaluate the performance of deep learning

models. The curve is typically depicted in a coor-

dinate system where the x-axis is the recall and the

y-axis is the precision and both values are always be-

tween 0 and 1. The graph shows how both metrics

change as the threshold is adjusted. A higher curve

generally points out better model performance. Ad-

ditionally, the area underneath the curve is known as

the average precision (AP) (Glassner, 2021).

AP is an essential performance indicator for one

object class, whereas the mean average precision

(mAP) is used when detecting multiple object cate-

gories in an image. It is calculated by the sub metrics

confusion matrix, IoU, recall and precision. “For each

class k, we calculate the mAP across different IoU

thresholds, and the ﬁnal metric mAP across test data

is calculated by taking an average of all mAP values

per class.” (Shah, 2022). Higher values show better

performance in object detection tasks. The mAP is

deﬁned as:

mAP =

k=n

∑

k=1

where AP

denotes the average precision of class k

and n is the number of classes. In subsequent sections

we use mAP@.5 to compare the outcome of the ex-

periments, which means a single IoU-threshold of 0.5

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

664

was applied. The mAP@.5 value seemed appropriate

for the use case, since counting is more essential than

precise localisation.

3.4 Training, Validation and Test

Model training, validation, and testing occurred on

Google Colab. We installed YOLOv7 and its depen-

dencies on the Colab virtual machine and utilized pre-

trained COCO dataset weights. After adjusting the

necessary conﬁguration ﬁles, the model underwent

training and subsequent testing for mAP results. Ten-

sorBoard helped in monitoring the training process.

4 RESULTS

This study conducted a series of experiments aiming

to enhance the mAP at an IoU threshold of 0.5 by

reﬁning the workﬂow parameters. We established a

baseline using coconut palm trees as the primary ob-

ject class.

4.1 Baseline

Establishing a baseline is pivotal for contextualiz-

ing model performance and offering a point of refer-

ence for subsequent experiments(Nair, 2022). Given

Ghana’s reddish laterite soil, DALL-E simulated

these background colors. Focusing on the objective

of counting coconut palm trees, we initially consid-

ered this as the sole object class. We generated 300

synthetic images for training, featuring between 15 to

25 trees each, totaling approximately 6000 trees. Af-

ter 40 epochs of training using pre-trained YOLOv7

weights, we achieved a mAP@.5 value of 0.65, de-

tailed in the subsequent section.

4.2 Varying Background Colors

Healthy coconut trees possess vibrant green leaves

with patches of brown, with color variations inﬂu-

enced by health, age, and environment. Notably, West

Africa’s laterite soil presents a reddish hue, while sur-

rounding vegetation remains chieﬂy green.

To ascertain the optimal background texture or

dominant color yielding the highest results with

trained YOLO models, backgrounds for training and

validation were crafted using stable diffusion, utiliz-

ing DALL-E, an AI that translates natural language

descriptions into images. As illustrated in Table 1, the

prompts were: dense green vegetation, red soil (offer-

ing enhanced contrast with the trees), and a combina-

tion of the two.

Each experiment iteration followed a similar pro-

cedure. The column ”Size of BG pool” speciﬁes the

backgrounds available for test (T) and validation (V)

datasets. Palm trees, rotated and placed at random,

ranged between 15 to 25 per image. Models were

trained over 40 epochs with a batch size of 16.

To account for result variability, models were

thrice tested on data taken 25 meters above ground.

Tests utilized the optimal model from prior training

in terms of mAP on the validation set (Skelton, 2022)

with an object conﬁdence threshold set at 0.6.

Figure 1’s bar plot delineates the mAP@0.5 val-

ues, revealing that the green background, akin to ac-

tual drone imagery grounds, outperformed others and

thus was selected for subsequent experiments.

The detect command results for the green back-

ground indicate fewer false positives/negatives than

its counterparts. Raising the conﬁdence threshold to

0.71 further reﬁned results. A potential inference

from Figure 1 is that the model’s capacity to discern

palm trees improves substantially when the stark con-

trast of green trees against a red backdrop isn’t the

dominant visual cue, but rather the model has to learn

other features for detection.

Figure 1: mAP@.5 values for different BG.

4.3 Multiple Object Classes

While the initial approach focused solely on identify-

ing coconut palm trees, certain plants like okra and

weeds were mistakenly identiﬁed as coconut palm

trees. This experiment aimed to determine if incorpo-

rating these misidentiﬁed objects as separate classes

would enhance the mAP@0.5 accuracy for the pri-

mary object – the coconut palm trees. Table 2 lists the

variants.

Okra, a staple in Ghanaian cuisine, can grow up

to 2 metres tall and has a notable presence in Ghana,

ranking it among the top ten global okra producers.

Its distinct shape, as well as that of tree trunks, eases

the generation of training data. However, tree trunks

Coconut Palm Tree Counting on Drone Images with Deep Object Detection and Synthetic Training Data

665

Table 1: Prompts to create background with DALL-E.

BG Color Prompt Size of BG Pool

green ”vertical aerial drone picture of ghana farm grass” T:35 V:21

red ”vertical aerial drone pictures of ghana soil” T:35 V:21

mixed equal composition of a and b T:35 V:21

Table 2: Set of object classes.

Abbr. Obj. Classes Cut out obj. Labelled (25m)

P. palm trees 44 187

P.O. palm trees, okra plants 44, 30 187, 2471

P.O.W.T. palm trees, okra plants, weeds, tree trunks 44, 30, 17, 24 187, 2471, 62, 193

Figure 2: Test results with one, two and four classes.

and patches of grass, which had prior misidentiﬁed,

were isolated and labelled. This necessitated extend-

ing the Python generator to accommodate multiple

object classes, as reﬂected in Table 2 with an aggre-

gate of 2, 911 plants.

Maintaining the green background, the workﬂow

remained consistent across trials. Results in Figure 2

suggest the four-class model offers superior mAP@.5

values. Furthermore, it demonstrates effective okra

recognition alongside the primary palm tree identiﬁ-

cation, achieving up to 85% accuracy for palm trees.

This highlights a potential strategy: reducing false

identiﬁcations in the primary object class by training

the model on frequently misidentiﬁed objects. How-

ever, this demands extensive manual labeling, as seen

with nearly 2, 000 okra plants in this case.

Subsequent experiments will utilize the green

background and four object classes, given their

demonstrated superiority in mAP@.5 outcomes.

4.4 Effect of Increasing Training

Images on Accuracy

This study investigates the potential correlation be-

tween increasing training images and the mAP@0.5

metric. The experiments conducted previously con-

sistently used 300 training and 120 validation images.

The most promising mAP@0.5 values emerged us-

ing the green BG and four object classes, which then

served as our consistent benchmark, as detailed in Ta-

ble 3.

Table 3: Number of training and validation images.

Training img Validation img ∅ Number palms

300 120 T:6000 V:2400

600 240 T:12000 V:4800

Notably, doubling the training images to 600 led

to a drop in mAP@0.5 values, not surpassing the 0.8

benchmark. This may suggest that using more images

from the same source pool can lead to model overﬁt-

ting, thus impairing its ability to generalize for unseen

data. This is further evidenced by the rise in false pos-

itives across all classes.

Furthermore, enlarging the training dataset corre-

spondingly extends the experiment duration, notably

during the upload to Google Drive and subsequent

Colab training.

Figure 3: Increased number of training images with 4

classes and green BG.

4.5 Inﬂuence of Drone Altitude on Test

Image Quality

Until now, training and validation data primarily came

from drone shots taken at approximately 10m and

25m above the ground. The test data was predomi-

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

666

nantly captured at 25m. With the most consistent re-

sults using the green BG, four object classes, and 300

training images, this conﬁguration became the base-

line for this study.

With the goal of counting all coconut trees using a

trained model, efﬁcient drone photography becomes

pivotal. A central question is determining the opti-

mal drone altitude to capture the maximum number

of trees without compromising image quality. While

greater altitude captures more trees in one shot, the

trees occupy fewer pixels, as detailed in Table 4.

Results, visualized in Figure 4, indicate that 70m

above ground provides the most promising test data.

However, due to the scarcity of high-altitude footage,

this conclusion serves only as an initial indicator.

While 70m seems to be the optimal altitude, this con-

clusion, based on only three test images with 66 co-

conut trees, requires further validation. Furthermore,

strategies must be developed to ensure comprehen-

sive land coverage with drone shots, avoiding gaps

or overlaps. Subsequent experiments will continue to

rely on the 25m data.

Table 4: Number of test images and palms in total.

Height of test images Image count Total palms

25m 38 187

45m 12 126

70m 3 66

Figure 4: Height of test images.

4.6 Number of Palm Trees on Training

Images

We investigated the impact of varying the number of

palm trees placed on backgrounds during training and

validation data generation. Prior experiments em-

ployed a random placement of 15 to 25 palm trees

per image. Table 5 enumerates the chosen ranges,

with results visualized in Figure 5a. It’s notable that

our model incorporated a green BG, identiﬁed the co-

conut palm tree as its single object class, and was

trained using 300 synthetic images.

A range of 5 to 15 palm trees yielded the highest

mAP@.5 scores. Given the average palm tree count

of 13 in the test drone footage, the model possibly

aligned better with test images having 2 to 14 palms.

This suggests that training and validation data with

similar palm tree ranges potentially improves model

assumptions on test data. On the contrary, the 15

to 25 range resulted in the lowest mAP@.5, possibly

due to the absence of this tree range in test images,

suggesting the model’s assumptions based on training

data might misalign. The 25 to 50 range saw mAP.5

improvements, likely reﬂecting the tree counts in test

data.

This theory warrants further exploration to better

understand model internalizations and their effects. A

subsequent experiment delves deeper into this.

Table 5: Number of palm trees in training and validation

images.

Palms per image ∅ Number of palms

5 to 15 3000

15 to 25 6000

25 to 50 11250

5 to 60 9750

4.7 Different Palm Count for Training

and Validation

Up to now the number range of palms in training and

validation data was equal. In the previous experiment

a theory was set up that there might be some internal-

isation in the model that the number of palm trees in

validation and test are similar to those in training. In

this section the range used for training and validation

data differs. Figure 5b depicts 3 variants, the ﬁrst two

variants with different ranges and the last variant with

same ranges for comparison. This is summarized in

table 6. The experiment explores if there is an im-

provement in the mAP.5 values by having different

training and validation ranges utilised for the training

of the model. It was trained with a green BG, one

object class (coconut palm trees) and 300 synthetic

images.

The ﬁrst variant shows slightly higher mAP@.5

values, which makes it one of the best results so far.

The second variant depicts lower mAP@.5 values.

The third variant show the result of the previous sec-

tion.

One possible explanation for the ﬁrst variant could

be that by training the model with different training

and validation ranges, it can handle the variety of

palm trees in each test image better. That means the

Coconut Palm Tree Counting on Drone Images with Deep Object Detection and Synthetic Training Data

667

(a) Number of palm trees (b) Range of palm trees

Figure 5: Numbers and range of palm trees in training and validation.

detection and classiﬁcation of palm trees are more ac-

curate. Another reason could be that the total number

of palms and their split in training and validation is

more important than the values per image.

Table 6: Number of palm trees in training and validation

images.

Palms per image ∅ Palms (train) ∅ Palms (val)

t:5-15 v:15-25 3000 2400

t:15-25 v:5-15 6000 1200

t:5-15 v:5-15 3000 1200

4.8 Freezing Layers

Yolov7, as detailed in section 2.4, boasts various lay-

ers with its initial weights trained on the Coco dataset.

When we freeze a layer, we prevent its weights from

updating during training. Typically, the architecture’s

initial layers capture basic features like edges. The

backbone of YOLOv7 encompasses 50 layers. In

this experiment, we adjusted the freeze hyperparam-

eter for ﬁne-tuning. Figure 6 presents outcomes when

ﬁxing the ﬁrst 5 and 11 layers. We trained using a

green BG, four object classes, and 300 images over 5

repetitions. The mAP@.5 values remained consistent

across both frozen and non-frozen variants, suggest-

ing that the freeze hyperparameter might not provide

added beneﬁts. Detection results revealed a high true

positive rate: of the 187 palm trees labeled in ground

truth, 199 were detected with minimal false positives.

4.9 Agricultural Implications

Through the conducted experiments, we successfully

elevated mAP@0.5 from an initial 0.65 to an average

of 0.88. For agricultural planning, such as procur-

ing fertilizers and protective nets, this accuracy proves

sufﬁcient. Table 7 consolidates the baseline and op-

timal variants from each experiment. Notably, in the

Figure 6: Freezing initial layers from 0-4 and 0-10.

ﬁnal freeze experiment, out of 187 labeled palm trees

in the ground truth, 199 were accurately detected with

minimal false positives. While all experiments in-

clude interpretations, further testing is essential for

validation, given the inherent variability in mAP.5 re-

sults even under consistent parameters. Enhancing

model transparency and clarity can also beneﬁt from

advances in explainable AI.

5 CONCLUSION

This research tackled the challenge of counting co-

conut palm trees in drone imagery using deep learn-

ing. Among various real-time object detectors,

YOLOv7 from the YOLO family emerged as the pre-

ferred choice. Given the limited availability of drone

footage, we generated synthetic images to train and

validate our model, which is also an important method

for this science. As a result, we witnessed a signiﬁ-

cant enhancement in the mAP@.5 value, elevating it

from 0.65 to 0.88. We varied input parameters and

ﬁne-tuned hyperparameters, ﬁnding that the genera-

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

668

Table 7: Comparison of best variants of selected experiments.

BG Classes Train Img Palms per Img Freeze Min mAP Max mAP

Red Palms 300 15-25 none 0.63 0.69

Green Palms 300 15-25 none 0.73 0.80

Green Palms 300 T:5-15 V:15-25 none 0.79 0.84

Green Palms 300 5-15 none. 0.69 0.86

Green P.O.W.T. 300 15-25 none 0.77 0.87

Green P.O.W.T. 300 5-15 0-10 0.75 0.88

tion of synthetic images, particularly with stable dif-

fusion backgrounds, was beneﬁcial. Our best model

detected 199 palm trees out of 187 labelled in the test

data, with minimal false positives.

For comprehensive results, systematic drone cap-

ture of land remains essential. Additional ﬁne-tuning

and experiments can further optimize the model.

Transitioning this methodology into a scalable prod-

uct holds potential for aiding nearby farms in yield

estimation and strategic planning. Future research

could also focus on assessing the health of coconut

palms. By integrating traditional farming with ad-

vanced techniques, we aimed to transform manual

surveys into a semi-automated, cloud-based solution.

This approach is not only cost-effective but also re-

duces time, labor, and errors. Bridging agriculture

with technology, especially through drones and deep

learning, unveils a horizon of promising opportuni-

ties.

ACKNOWLEDGEMENTS

This paper was partially funded by the German Fed-

eral Ministry of Education and Research through the

funding program ”quantum technologies - from basic

research to market” (contract number: 13N16196).

Furthermore, this paper was also partially funded

by the German Federal Ministry for Economic Af-

fairs and Climate Action through the funding program

”Quantum Computing – Applications for the indus-

try” (contract number: 01MQ22008A).

REFERENCES

Anwer, A. (2022). What is average precision in object de-

tection and localization algorithms and how to calcu-

late it?

Glassner, A. (2021). Deep Learning: A Visual Approach.

No Starch Press, San Francisco, California, 1st edi-

tion.

Goodfellow I., Bengio Y., C. A. (2016). Deep Learn-

ing (Adaptive Computation and Machine Learning se-

ries). The MIT Press, Cambridge, Massachusetts, 1st

edition.

Khandelwal, R. (2020). Tensorﬂow object detection api.

Kukil and Rath, S. (2022). Yolov7 object detection paper

explanation and inference.

Nair, A. (2022). Baseline models: Your guide for model

building.

Prince, S. J. (2023). Understanding Deep Learning. MIT

Press, Cambridge, Massachusetts.

Redmon, J., Divvala, S., Girshick, R., and Farhadi, A.

(2016). You only look once: Uniﬁed, real-time object

detection. In Proceedings of the IEEE conference on

computer vision and pattern recognition, pages 779–

788.

Sevarac, Z. (2021). How to get started with deep learning

in java.

Shah, D. (2022). Mean average precision (map) explained:

Everything you need to know.

Skelton, J. (2022). Step-by-step instructions for training

yolov7 on a custom dataset.

Szeliski, R. (2022). Computer Vision: Algorithms and Ap-

plications. Springer, Cham, Switzerland, 2nd edition.

Coconut Palm Tree Counting on Drone Images with Deep Object Detection and Synthetic Training Data

669