Mobile Application with Convolutional Neural Networks for the

Early Detection of Diseases in Blueberry Plants in Chepén: Trujillo

Santiago Sebastian Heredia Orejuela and Aaron Moises Cosquillo Garay

School of Computer Engineering, Universidad Peruana de Ciencias Aplicadas (UPC), Lima, Peru

Keywords: Blueberry Plants, Convolutional Neural Networks, Mobile Application, Disease Detection, Offline Diagnostics.

Abstract: Early detection of foliar diseases in blueberry crops is essential to protect yield and fruit quality, especially in

Chepén–Trujillo, a key agricultural region in Peru. This paper presents a mobile application developed with

Flutter and powered by a lightweight convolutional neural network (CNN), capable of analyzing leaf images

and delivering disease diagnoses in under three seconds. The system supports offline functionality, ensuring

usability in rural areas with limited connectivity. In a test set of 350 images, the model achieved 93% accuracy,

88% recall, and an F1 score of 0.90. Field validation with local farmers showed 90% agreement with expert

diagnoses. Beyond its technical performance, this solution has the potential to reduce economic losses,

improve crop quality, and empower smallholder farmers through accessible, real-time diagnostics. The

platform is scalable to other crops and regions, contributing to more sustainable and resilient agricultural

practices in Peru.

1 INTRODUCTION

Blueberry production in the Chepén–Trujillo region

plays a strategic role in the local economy, not only

due to its relevance in national and international

markets, but also for its contribution to rural

development. In recent years, Peru has become the

world’s leading exporter of blueberries, generating

over $1 million in revenue in 2023 alone (Plataforma

del Estado Peruano, 2023). However, this success is

increasingly threatened by the late detection of plant

diseases, which remains one of the main causes of

yield losses and reduced fruit quality. This issue is

further exacerbated by limited connectivity in

agricultural areas, which prevents real-time access to

advanced technological tools and expert diagnostics.

In this context, the integration of artificial

intelligence into agriculture offers a promising

solution. Convolutional neural networks (CNNs)

have demonstrated high accuracy in the automated

identification of crop diseases through image analysis

(Ferentinos, 2018; Bhuvana & Mirnalinee, 2021).

These models significantly reduce false positives and

false negatives, improving the speed and reliability of

disease diagnosis. However, most existing CNN-

based systems are developed in controlled

environments and are not adapted to the specific

conditions of blueberry cultivation in Chepén–

Trujillo, such as variable lighting, dust interference,

and limited internet access.

This research hypothesizes that a mobile

application powered by a lightweight CNN model can

significantly improve the early detection of foliar

diseases in blueberry plants, thereby reducing

production costs and increasing crop productivity.

The proposed solution is a Flutter-based mobile

application that allows farmers to capture images of

blueberry leaves, preprocess them on the device, and

obtain real-time diagnostic results through an

embedded CNN or a RESTful API. The system also

supports offline queuing and synchronization,

ensuring usability in low-connectivity environments.

By providing an accessible, portable, and efficient

tool for disease detection, this project aims to

empower smallholder farmers, reduce economic

losses, and enhance the sustainability of one of Peru’s

most important agricultural exports.

2 RELATED WORKS

This literature review was developed to address the

key research questions in our framework: the

effectiveness of CNN-based models for plant disease

192

Orejuela, S. S. H. and Garay, A. M. C.

Mobile Application with Convolutional Neural Networks for the Early Detection of Diseases in Blueberry Plants in Chepén: Trujillo.

DOI: 10.5220/0013676800003982

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

In Proceedings of the 22nd International Conference on Informatics in Control, Automation and Robotics (ICINCO 2025) - Volume 1, pages 192-199

ISBN: 978-989-758-770-2; ISSN: 2184-2809

detection, their integration into mobile applications,

and the challenges of deploying these models in real-

world rural environments such as Chepén–Trujillo.

Sources were identified through a structured search in

academic databases including IEEE Xplore,

ScienceDirect, and Scopus, using keywords such as

"plant disease detection," "convolutional neural

networks," "mobile application," and "agricultural

deep learning." Articles were selected based on their

methodological quality, reported performance

metrics, and applicability to the constraints of low-

connectivity agricultural regions.

Convolutional Neural Networks (CNNs) have

consistently demonstrated superior performance in

image-based plant disease detection. For example,

PDDNetcv, proposed by Bhuvana & Mirnalinee

(2021), was able to classify 15 different leaf diseases

with a mean average precision of 99.09% and

achieved high F1-scores across multiple classes.

Similarly, Ferentinos (2018) reported accuracy levels

above 99.5% on models trained with the PlantVillage

dataset, thanks to aggressive data augmentation

strategies and hyperparameter tuning. These studies

validate the capability of CNNs to learn intricate

visual features under diverse lighting and background

conditions, which is critical for field applications.

Several works have also explored the deployment

of CNN models into real-time mobile applications.

Janarthan et al. (2022) developed an ultra-lightweight

(1 MB) offline mobile app for palm disease detection,

enabling use in areas without reliable internet.

Waheed et al. (2023) combined MobileNetV2 and

VGG16 to construct a mobile tool for ginger leaf

analysis, achieving approximately 99% classification

accuracy. Meanwhile, Lanjewar and Parab (2024)

utilized transfer learning techniques to implement a

citrus disease classifier hosted on a Platform-as-a-

Service (PaaS) system, showcasing the feasibility of

cloud-trained, smartphone-executed models in

agriculture.

Moreover, newer architectures such as

EfficientNet and Vision Transformers have been

explored to improve performance while reducing

computational costs. Tapia et al. (2024) and

Valenzuela (2021) showed that these models can

achieve over 90% accuracy on tasks ranging from

tomato leaf diagnosis to satellite-based crop

monitoring, reinforcing their versatility across

datasets. However, these models typically require

high-resolution labeled datasets and stable

infrastructure, which may not align with the realities

faced by smallholder farmers in Peru.

Despite the advances, a critical research gap

remains none of the reviewed studies address the

detection of blueberry leaf diseases under the specific

conditions found in Chepén–Trujillo namely, varying

lighting conditions, dust interference, and

intermittent mobile data. While blueberry cultivation

has gained significant traction in Peru's northern

regions, most current AI tools remain generalized or

crop-agnostic, thereby limiting their diagnostic

reliability in real-world field scenarios.

In response, our work focuses on the design and

implementation of a mobile solution tailored

specifically to blueberry disease detection. The model

was trained on a cloud platform using 32 and 64

image batch sizes, an initial learning rate of 0.001,

and learning rate decay to optimize convergence.

During experiments with a test set of 350 images, our

CNN model achieved a precision of 93%, recall of

88%, and an F1-score of 0.90. These performance

metrics suggest robust classification performance

across key blueberry disease categories such as

anthracnose, rust, and bacterial spots.

Furthermore, real-world validation was conducted

through field tests in collaboration with local farmers.

The model’s predictions showed a 90% agreement

with expert agronomist diagnoses, confirming its

potential as a practical tool for non-specialist end-

users in rural contexts. These results position our

work as a pioneering application in the Peruvian

agricultural tech landscape.

Lastly, while many mobile applications rely on

cloud connectivity for inference, our solution was

explicitly designed to function offline, addressing the

connectivity challenges frequently encountered in

agricultural zones. By maintaining lightweight

architecture and storing a history of previous

diagnoses on the device, the system ensures both

resilience and usability in the field.

3 METHODS

This research follows a structured methodology

divided into seven key stages, ensuring both technical

rigor and practical applicability:

3.1 Data Acquisition and Preparation

Images were sourced from two main origins: the

PlantVillage public dataset, known for its diversity of

labeled plant disease images, and field photos taken

at Frutas del Norte S.A.C. farms in Chepén–Trujillo,

Peru. This dual-source approach ensured both

generalization and local relevance. All images were

manually annotated by plant pathology experts,

Mobile Application with Convolutional Neural Networks for the Early Detection of Diseases in Blueberry Plants in Chepén: Trujillo

193

following a standardized taxonomy of foliar diseases

including anthracnose, rust, and bacterial leaf spots.

Preprocessing steps included image resizing to

224×224 pixels to ensure compatibility with the CNN

input layer, normalization of pixel values for

improved convergence and data augmentation

(rotation, scaling, horizontal flipping) to increase the

diversity of training examples and reduce overfitting.

The final dataset was partitioned into training

(70%), validation (15%), and testing (15%) sets using

stratified sampling to preserve class distribution. This

step was critical to maintain consistency in

performance metrics across disease classes.

3.2 CNN Model Development

A pre-trained MobileNetV2 architecture was selected

due to its lightweight design and proven accuracy in

mobile vision tasks. This model offers an optimal

trade-off between inference speed and classification

performance on resource-constrained devices.

Using transfer learning, the early convolutional

layers were frozen to retain pre-learned low-level

features. The following custom layers were appended

global average pooling to reduce dimensionality, two

fully connected (dense) layers for deeper feature

learning, a dropout layer (rate: 0.5) to prevent

overfitting and Softmax output layer for multi-class

classification.

ReLU activation functions were used in the dense

layers to introduce non-linearity and enhance learning

capacity.

3.3 Model Training and Validation

Training was conducted on Google Colab Pro,

utilizing GPU acceleration (Tesla T4) to speed up

computation. The model was compiled with batch

sizes of 32 and 64 (empirically tested), initial learning

rate of 0.001 with exponential decay, categorical

cross-entropy as the loss function and adam optimizer

for adaptive learning rate adjustments.

To avoid overfitting, early stopping and model

checkpointing were implemented, monitoring

validation accuracy and halting training when

performance plateaued. The best-performing model

(based on validation accuracy) was retained for

deployment.

3.4 Mobile Application Development

The application was developed using Flutter,

enabling cross-platform deployment on both Android

and iOS with a single codebase. The user interface

(UI) was designed to prioritize simplicity and

accessibility, incorporating real-time camera

integration and gallery upload, on-device predictions

diagnosis history and visual and audio feedback

mechanisms to support low-literacy users.

The interface was iteratively improved through

user feedback sessions with farmers in Chepén.

3.5 RESTful API Integration

In addition to local inference, a Flask-based REST

API was developed to support remote classification

when connectivity is available. The API accepts

POST requests containing encoded leaf images,

returns classification results with confidence scores,

utilizes token-based authentication to ensure secure

communication.

This hybrid design enables both offline and online

modes, adapting to varying field conditions.

3.6 Evaluation and Testing

The model was evaluated using a test set of 350 leaf

images. It achieved an accuracy of 93%, a recall of

88%, and an F1 score of 0.90. These results indicate

that the model is highly effective in correctly

identifying both healthy and diseased blueberry

leaves, with a strong balance between precision and

recall. Additionally, field testing was conducted in

collaboration with local farmers in the Chepén-

Trujillo region, where real-world performance was

assessed. The system’s diagnostic results showed a

90% agreement with those of agricultural experts,

confirming its potential for practical deployment in

farming contexts.

To assess usability, several human-centered

metrics were collected through structured interviews

and in-app surveys. The task success rate defined as

the ability of users to complete a diagnosis without

assistance was above 92%, while error frequency

remained low, primarily related to poor image quality

or inconsistent lighting. Furthermore, user

satisfaction was rated positively, with farmers

highlighting the ease of use, quick response time, and

potential for reducing crop loss through early disease

detection.

Overall, the model not only demonstrated strong

technical performance but also showed promise as a

reliable and accessible tool for supporting decision-

making in agricultural practices. These findings

suggest that integrating AI-based solutions into

mobile platforms can significantly enhance disease

management in crops like blueberries, especially in

ICINCO 2025 - 22nd International Conference on Informatics in Control, Automation and Robotics

194

regions with limited access to agricultural specialists.

3.7 Project Management and Planning

The development process adhered to the Scrum

framework, with three-week sprints, daily stand-ups,

sprint reviews, and retrospectives. PMBOK

principles were incorporated for documentation, risk

management, and quality control. The project

timeline included three phases: planning and research

(2 weeks), development and training (8 weeks), and

testing and deployment (4 weeks).

4 EXPECTED RESULTS

This project aims to deliver a practical and impactful

solution for blueberry farmers in Chepén by

achieving the following outcomes:

4.1 Technical Performance in Local

Conditions

Accurate Disease Detection in Field and

Laboratory Conditions: The convolutional neural

network (CNN) model was trained on a curated

dataset of blueberry leaf diseases, including both

public data (PlantVillage) and field images from

Chepén. During model training on a cloud platform

using batch sizes of 32 and 64, an initial learning rate

of 0.001 was employed with exponential decay. The

model architecture was based on MobileNetV2 with

transfer learning, optimized for mobile inference

using TensorFlow Lite.

In laboratory-controlled experiments, the model

was evaluated using a separate test set of 350 labeled

images. The following metrics were obtained:

Table 1: Metrics of model.

Metric Value

Accurac

93%

Recall 88%

F1 Score 0.90

These results indicate strong discriminative

power, particularly in distinguishing early-stage

disease symptoms from similar-looking healthy

leaves or non-disease stress conditions.

To validate real-world applicability, field testing

was conducted in collaboration with local farmers.

The app's diagnoses were compared against those

made by certified agronomists, showing a 90%

agreement rate. This demonstrates the robustness of

the model under variable lighting, growth stages, and

mobile device constraints typical of agricultural

environments in Chepén.

To further analyze the model’s diagnostic

capability, we evaluated its performance across

different blueberry diseases. Table 2 summarizes the

accuracy, recall, and F1-score for three representative

diseases included in the dataset.

Table 2: Classification performance by disease type.

Disease

Accuracy

(%)

Recall

(%)

F1-

score

Anthracnose 95 91 0.93

Rus

92 88 0.90

Bacterial leaf

spo

90 84 0.87

These results confirm that the model performs

reliably across various foliar diseases, despite

differences in visual symptoms and data imbalance.

4.2 Real-Time and Offline Diagnosis

The mobile application has been specifically

optimized to run efficiently on low-spec Android

devices, which are widely used in rural and

agricultural regions of Peru, particularly in areas like

Chepén and Trujillo. This optimization ensures that

farmers and field workers can use the tool without

needing high-end smartphones, making the

technology more accessible and inclusive for a

broader range of users, regardless of their

socioeconomic status or digital infrastructure.

During field testing, the application demonstrated

the ability to deliver real-time, on-device diagnostic

results within seconds of capturing a leaf image. This

immediate feedback is critical in agricultural

environments where timely decisions can

significantly influence the effectiveness of disease

control measures and ultimately impact crop yield.

The real-time analysis is powered by a lightweight

convolutional neural network (CNN) model that has

been trained and quantized for mobile deployment,

allowing it to run efficiently without relying on cloud-

based processing.

One of the key strengths of the system is its robust

offline functionality. Recognizing the limited and

often intermittent internet connectivity in rural areas,

the application has been designed to operate entirely

offline for core functions such as image capture,

disease detection, and local storage of results. Users

can perform multiple diagnoses throughout the day

without requiring any connection to external servers.

This design choice not only ensures uninterrupted

Mobile Application with Convolutional Neural Networks for the Early Detection of Diseases in Blueberry Plants in Chepén: Trujillo

195

usability in the field but also helps reduce data costs

for users.

When an internet connection becomes available

whether through Wi-Fi in a nearby town or mobile

data the application automatically synchronizes

locally stored data with a centralized server. This

synchronization process uploads diagnostic results,

usage logs, and images, which can then be accessed

by agronomists or researchers through a web-based

dashboard for further analysis and monitoring. It also

allows for the aggregation of disease data across

regions, which can contribute to larger-scale studies

and early warning systems for disease outbreaks.

Moreover, the asynchronous design of the data

synchronization process ensures that users are not

affected by delays or interruptions during app usage.

This hybrid real-time/offline approach balances

performance, reliability, and accessibility, making the

application a practical and valuable tool for

smallholder farmers and agricultural technicians

working in low-connectivity environments.

4.3 Accessibility and Usability for

Local Farmers

The application design prioritizes user-centered

interaction for rural and low-digital-literacy users.

Based on user testing sessions with local farmers, the

interface was simplified using icons, minimal text,

and step-by-step guidance. Feedback collected

through structured interviews was used iteratively to

improve usability.

Additionally, the app includes voice instructions

in Spanish, tooltips, and a minimal learning curve for

first-time users. These features are essential for

scaling adoption in rural agricultural communities

with diverse technological access levels.

4.4 Sustainable and Scalable

Agricultural Impact

Contribution to Sustainable Farming Practices:

The proposed application plays a significant role in

promoting sustainable agriculture by enabling early

and accurate detection of plant diseases. This early

intervention capability allows farmers to apply

treatments only when and where they are truly

needed, significantly reducing the indiscriminate use

of chemical pesticides. As a result, the platform helps

to lower the environmental footprint associated with

traditional farming practices, particularly in terms of

soil and water contamination. Additionally, by

avoiding unnecessary agrochemical applications,

farmers can decrease their operational costs while

simultaneously enhancing the overall health and

productivity of their crops. In the long term, this can

contribute to more resilient and eco-friendly

agricultural systems, aligned with global

sustainability goals such as those set by the United

Nations (e.g., SDG 2: Zero Hunger and SDG 12:

Responsible Consumption and Production).

Scalability to Other Crops: Although the current

convolutional neural network (CNN) model has been

specifically trained and optimized for disease detection

in Vaccinium corymbosum (blueberry), the system has

been deliberately designed with scalability in mind. Its

modular and flexible architecture allows for the

integration of new classification models with minimal

disruption to the core application. By collecting and

annotating image datasets for other crops, such as Vitis

vinifera (grape) and Persea americana (avocado), the

application can be adapted to meet the needs of

different agricultural sectors.

These crops have been selected not only because

of their economic relevance in the Chepén–Trujillo

region, but also due to their susceptibility to various

fungal and bacterial diseases that, if detected early,

can be effectively controlled. The training pipeline

supports transfer learning techniques, which allows

the system to leverage knowledge from existing

models and accelerate the development of new crop-

specific detectors.

A future update of the platform will include a

dynamic model selection feature within the app’s

interface, enabling users to switch between crop types

according to their specific cultivation needs. This

functionality ensures that the application remains

useful and adaptable across different contexts and

agricultural cycles, ultimately increasing its adoption

potential among diverse user groups.

Long-Term Vision: By continually expanding the

app’s capabilities and supported crops, the system has

the potential to evolve into a comprehensive

agricultural assistant. Future iterations could integrate

features such as nutrient deficiency analysis, and

integration with weather prediction models for more

holistic crop management. Such enhancements would

further reinforce the platform's value as a sustainable

and scalable solution for precision agriculture,

particularly in regions where access to expert

agronomic support is limited.

4.5 Interpretation and Comparison of

Results

The results obtained in the laboratory and field tests

not only show a high performance of the CNN model

ICINCO 2025 - 22nd International Conference on Informatics in Control, Automation and Robotics

196

but also allow us to reflect on its behaviour in real

contexts versus ideal conditions.

In the controlled environment, the model achieved

outstanding metrics (93% accuracy, 88% recall and

F1-score of 0.90), confirming its ability to

discriminate between healthy and diseased leaves

with high reliability. However, it is in field tests that

its true practical usefulness is put to the test. The 90%

agreement with expert agronomists' diagnoses

demonstrates that the system maintains robust

performance even in the face of variations in lighting,

image quality and environmental conditions.

This 3% difference between lab accuracy and

field agreement can be attributed to factors such as

visual noise in images captured by farmers

(background, shadows, dust), variability in the mobile

devices used (camera resolution, processing) and

different phenological conditions of the plants do not

present in the training set.

Despite these limitations, the model showed

adequate generalizability, which validates its design

and training. Furthermore, the fact that it works

without internet connection and on mid-range devices

reinforces its applicability in rural areas with limited

infrastructure.

Taken together, these results not only validate the

technical architecture of the system but also support

its viability as a real-time agricultural decision

support tool.

5 DISCUSSIONS

The successful implementation of this project has

the potential to significantly improve disease

management practices in blueberry farming.

Comparative Evaluation with Other Models

To contextualize the model’s performance, we

compared it with other state-of-the-art architectures.

Table 3 presents a benchmark between MobileNetV2

(our selected model), EfficientNetB0, and traditional

agronomist diagnosis methods.

MobileNetV2 strikes a balance between high

accuracy and low latency, making it ideal for mobile

applications in rural environments. While

EfficientNetB0 offers comparable accuracy, its larger

size and longer inference time limit its usability on

low-spec devices. Traditional manual diagnosis,

although accurate, lacks speed and scalability,

especially in remote areas.

Table 3: Comparative evaluation of disease detection

approaches.

Model /

Method

Accura

cy (%)

Infere

nce

Time

(s)

Model

Size

(MB)

Offline

Capabil

ity

MobileNetV

2 (ours)

93 2.8 14 Yes

EfficientNet

91.5 5.9 20 Partial

Agronomist

diagnosis

(avg)

Manu

al (1–

2 min)

- -

The use of CNNs offers a more accurate and

efficient alternative to traditional manual inspection

methods. The mobile application will empower

farmers to proactively identify and address diseases,

reducing crop losses and improving overall yield.

However, several challenges need to be addressed

to ensure the project's success. The quality and

diversity of the training data are critical to the model's

performance. Careful attention must be paid to data

collection, labeling, and augmentation. The

computational resources required for training and

deploying the CNN model must be carefully

managed. The application's user interface must be

designed to meet the specific needs and technical

skills of the target users.

Limitations and Critical Reflection

Despite its promising results, the proposed system

faces several limitations. One key concern is the bias

in the dataset, which was limited in representing rare

diseases or extreme leaf conditions. This may lead to

misclassifications, particularly in cases where

multiple symptoms overlap or visual noise (e.g.,

shadows, dust) is present in the captured image.

Another limitation is the scalability of the solution

to other regions and crops. While the modular design

allows model retraining, performance may vary

depending on environmental conditions, device

quality, and farmer practices.

Errors in classification were most common in

early-stage diseases with subtle symptoms. Future

versions should incorporate active learning or user

feedback to refine predictions over time.

Additionally, the app’s offline capabilities,

although essential, limit the ability to store large

datasets locally or receive real-time updates, which

could affect long-term adaptability.

The integration of Scrum and PMBOK

methodologies will provide a structured and

adaptable framework for managing the project's

complexity. Regular communication with

Mobile Application with Convolutional Neural Networks for the Early Detection of Diseases in Blueberry Plants in Chepén: Trujillo

197

stakeholders, including farmers and agricultural

experts, will be essential to ensure that the project

remains aligned with their needs and expectations.

6 CONCLUSIONS

This project aims to develop a mobile application that

leverages the power of CNNs to provide early and

accurate disease detection in blueberry plants. The

project's success will depend on careful data

management, robust model development, user-

centered design, and effective project management.

The expected outcomes of this project include a

high-accuracy disease detection model, a user-

friendly mobile application, and scalable architecture.

The successful deployment of this application has the

potential to transform disease management practices

into blueberry farming, improving crop yields,

reducing losses, and promoting sustainable

agricultural practices. The project's findings will

contribute to the growing body of knowledge on the

application of AI and deep learning in agriculture.

Further research and development efforts will be

focused on expanding the application's functionality

to include other crops and regions, improving the

model's accuracy and robustness, and integrating the

application with other agricultural management tools.

In future iterations, we aim to expand the system’s

functionality through multimodal disease prediction,

integrating data from environmental sensors such as

soil moisture, temperature, and humidity to improve

diagnostic accuracy. Additionally, we plan to train

new models on aerial images captured by drones,

which would allow broader monitoring of crop health

at the plantation level.

Furthermore, we intend to assess the long-term

socio-economic impact of the mobile application,

particularly regarding its influence on decision-

making, pesticide usage, and smallholder farmers’

income. These developments will support a more

holistic and sustainable approach to agricultural

disease management in rural Peru.

ACKNOWLEDGMENTS

First, we would like to thank the Universidad Peruana

de Ciencias Aplicadas (UPC) for providing us with

the academic training and resources necessary to

develop this project. To our teachers and advisors,

who with their guidance, knowledge and constant

academic support, contributed significantly to the

development of this research.

Special thanks to the farmers of Chepén-Trujillo,

who actively participated in the field trials and shared

their experience and knowledge of blueberry

cultivation. Their collaboration was fundamental to

validate the practical usefulness of the application

developed.

To the company Frutas del Norte S.A.C., for

allowing access to their fields for data collection and

for their willingness to collaborate with the research.

To my family, for their unconditional support,

patience and constant motivation throughout this

process. Their confidence in us has been a

fundamental pillar to reach this goal.

REFERENCES

Bhuvana, J., & Mirnalinee, T. (2021). An approach to Plant

Disease Detection using Deep Learning Techniques.

Iteckne, 18(2), 161–169. https://doi.org/10.15332/

iteckne.v18i2.2615

Ugail, h. (2022). deep learning in visual computing:

explanations and examples. CRC Press.

https://goo.su/tAlZf0

Janarthan, S., Thuseethan, S., Rajasegarar, S., & Yearwood,

J. (2022). P2OP—Plant Pathology on Palms: A deep

learning-based mobile solution for in-field plant disease

detection. Computers and Electronics in Agriculture,

202, 107371 https://doi.org/10.1016/j.compag.

2022.107371

Waheed, H., Akram, W., Islam, S., Hadi, A., Boudjadar, J.,

& Zafar, N. (2023). A Mobile-Based System for

Detecting Ginger Leaf Disorders Using Deep Learning.

Future Internet, 15(3), 86. https://doi.org/10.3390/

fi15030086

Lanjewar, M., & Parab, J. (2024). CNN and transfer

learning methods with augmentation for citrus leaf

diseases detection using PaaS cloud on mobile.

Multimedia Tools and Applications, 83(11), 31733–

31758. https://doi.org/10.1007/s11042-023-16886-6

Tapia, J., Castro, G., Villanueva, J., & Castro, L. (2024).

Detección de enfermedades y plagas en cultivos de

tomate mediante el análisis de imágenes con Deep

Learning. [Tesis de maestría, Universidad Peruana de

Ciencias Aplicadas]. Repositorio Académico UPC.

https://upc.aws.openrepository.com/handle/10757/675

351

Valenzuela, S. (2021). Detección y clasificación de

enfermedades en el tomate mediante Deep Learning y

Computer Vision. [Tesis de magister, Universidad

Nacional de La Plata, Facultad de Ingeniería. La Plata,

Argentina]. Repositorio Institucional de la UNPL

https://sedici.unlp.edu.ar/handle/10915/139770

Gamez, K. (2023). Deep Learning en agricultura: conceptos

y aplicaciones en la identificación de cultivos sobre

imágenes satelitales. [Tesis de título, Universidad de los

ICINCO 2025 - 22nd International Conference on Informatics in Control, Automation and Robotics

198

Andes, Facultad de Ingeniería. Bogotá, Colombia].

Repositorio Institucional UNIANDES https://hdl.

handle.net/1992/73839

Ferentinos, K. (2018). Deep learning models for plant

disease detection and diagnosis. Computers and

Electronics in Agriculture, 145, 311–318.

https://doi.org/10.1016/j.compag.2018.01.009

Plataforma del Estado Peruano (2023). Perú, por

cuarto año consecutivo, vuelve a ser el primer

exportador mundial de arándanos. Gob.pe.

Recuperado de https://www.gob.pe/institucion/

embajada-del-peru-en-espana/noticias/737763-peru-

por-cuarto-ano-consecutivo-vuelve-a-ser-el-primer-

exportador-mundial-de-arandanos.

Mobile Application with Convolutional Neural Networks for the Early Detection of Diseases in Blueberry Plants in Chepén: Trujillo

199