Efficient Detection of Cucurbit Pepo Leaf Diseases Using Advanced

Image Processing Techniques

P. A. Selvaraj

, K. Dhanushree

, G. K. Ranga Kaarthi

and R. K. Sanjith

Department of CSD, Kongu Engineering College, Tamil Nadu, India

Keywords: Curcurbita Pepo, Leaf Disease Detection, YOLOv7, Convolutional Neural Networks, CNN, Precision

Agriculture, Computer Vision, Plant Pathology, Disease Classification, Agricultural Research, Pumpkin.

Abstract: Curcurbita pepo leaf diseases are among the critical factors that bring down agricultural productivity and,

therefore, require an accurate diagnostic tool. In this paper, there has been proposed a classification model for

recognizing four common diseases of Cucurbita pepo leaf—Downy Mildew, Powdery Mildew, Mosaic

Disease, and Bacterial Leaf Spot—together with healthy leaves using YOLOv7 and Convolutional Neural

Networks. In this paper, the authors use the Cucurbita pepo leaf disease dataset, which includes 2000 high-

resolution images, to correctly classify a given leaf as healthy or infected. The dataset is well structured and

will enhance studies investigating disease symptoms, becoming very useful in agricultural studies and

education. Our findings demonstrate how state-of-the-art computer vision models could be put into practice

to improve disease diagnosis for the promotion of precision agriculture and automated systems for real-time

monitoring of diseases at the point of intervention. Therefore, such technologies will help the agricultural

sector realize efficient management of diseases since reduced losses will result in higher quality yields. This

research highlights the realization of incorporating artificial intelligence and machine learning in farming

processes to minimize challenges and improve productivity.

1 INTRODUCTION

The agricultural sector is important due to its

contribution to world economic growth and the

supply of food to all human beings. On the other

hand, with the increased deterioration of the

environment, there has been a high increase in

diseases that attack plants. Pathogens that invade

plants may cause losses of production in agricultural

produce. It's time-consuming and difficult to inspect

trees for disease presence and it needs a lot of human

and financial resources (Bonkra, Pathak, et al. , 2024).

About 160,000 hectares of land in the valley are under

horticultural cultivation. This is followed by the

production of around 180,000 metric tons annually.

According to the Horticulture Division, 2021, this

forms a significant share of exports across the world

over the recent past. However, annual losses incurred

https://orcid.org/0009-0002-7278-8688

https://orcid.org/0009-0009-0505-5848

https://orcid.org/0009-0005-4028-6327

https://orcid.org/0009-0004-1821-5461

by these pests and diseases are very huge in the fruit

juice industry. The diseases that still haunt the apple

growers are Alternaria, scab, and mosaic. In July

2013, Alternaria breakout was realized from the juice,

and it spread like wildfire, spreading over 70 percent

of the variety to different parts of the valley. (Khan,

Quadri, et al. , 2022) Across the world, pests and

diseases destroy an entire apple crop. In Himachal

Pradesh, the second largest apple producer in India,

one major cause that affects the quality of apples is

fungal disease. Plant diseases might be broadly

grouped into biotic and abiotic diseases. They are

disease-causing organisms, like bacteria, viruses, and

fungi (Galbraith, 2024). Compared to non-infectious

diseases, bacterial infections are common and

dangerous about physiological diseases such as

mineral deficiency, sunburn, and other environmental

factors. Some of the diseases found on the leaves

470

Selvaraj, P. A., Dhanushree, K., Kaarthi, G. K. R. and Sanjith, R. K.

Efﬁcient Detection of Cucurbit Pepo Leaf Diseases Using Advanced Image Processing Techniques.

DOI: 10.5220/0013622000004664

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

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 3, pages 470-475

ISBN: 978-989-758-763-4

include Downy mildew, Powdery mildew, Mosaic

disease, and Bacterial leaf spot (Vishnoi, Kumar, et

al., 2021).

2 RELATED WORKS

Many researchers have made efforts to investigate

early crop diseases. In this model, we focus on the

classification of squash disease and fruit disease and

use the combined model of neural network to identify

squash disease and fruit disease, and achieve better

identification of squash blight. This article presents

studies that illustrate many of the methods currently

used to identify plant and leaf diseases. A brief

description of these issues can be found here.

1) Downy mildew is here in Knox County. Thus

far, the disease has not been reported on squash, but

it is certainly possible for downy mildew to move

from cucumbers onto squash. Squash growers must

monitor for minor diseases and develop their control.

Squash growers who will harvest in mid-September

need to apply fungicide before the end of August.

This translates to just one or two more applications.

Because downy mildew does not affect fruit, there is

no need to apply fungicide before harvest. Not

anticipating the fruit to be growing until mid-October,

the grower decides to control it in September.

(Salcedo, Purayannur, et al. , 2021)

2) Powdery mildew - Usually caused by the

fungus Podosphaera xanthii, powdery mildew infects

all cucurbits, including melons, squash, cucumbers,

gourds, watermelons, and pumpkins. New spores can

form and spread easily in a warm, dry environment.

Older leaves are more susceptible to the disease and

powdery mildew will affect them first. On each page.

Although powdery mildew usually infects leaves

and vines, it can also infect cucumbers or melons.

Powdery mildew does not infect squash fruit directly.

Fruits do not do well due to excess sunlight, immature

ripening, instability, and Odor. Planted. While the

attack of powdery mildew is generally on the leaves

and vines, sometimes it also affects the cucumbers or

melons. (Landschoot, Abbey, et al. , 2024) In the case

of squash, powdery mildew does not actually infect

the fruit. Due to too much sunlight, immature

ripening, instability, and Odor, fruits do not come up

well.

3) Mosaic diseases are diseases that destroy

plants, gardens and crops at the molecular level.

When a plant is infected with mosaic virus, the

infected plant can spread the disease to other plants

and affect the entire crop if not treated quickly.

Mosaic disease can be spread by plants, infected

seeds, infected plants, or some insects. Aphids,

grasshoppers, mealybugs, and cucumber beetles are

common garden pests that can spread this disease.

Aphids are the most common insects in the garden, so

understanding aphid control is crucial to any garden

or harvest (Vinje, 2024) Contaminated soil, seeds,

fermenters and containers will also become infected

and transmit the disease to the plant. Cutting or

splitting the tissue can carry the virus and cause it to

spread.

4) Leaf spot disease is one of the most important

diseases of cucurbits and affects almost all crops

worldwide. The disease has been reported to cause

significant losses in cucurbit plants because the

symptoms appear on all parts of the plant, including

the fruit. This disease is suitable for temperate and

cold conditions. The virus persists in the seed and the

crop. Efforts to control the disease are ongoing with

reports of over 50% control in the field. There is

currently not much known to be resistant to this

disease, but it can be controlled with antibiotics.

(Jarial, Jarial, et al. , 2023)

Controlling giant squash disease in the home

garden begins with planting resistant varieties and

using good cultural practices. New varieties resistant

to diseases such as powdery mildew and fusarium

fruit rot are introduced each year. Whenever possible,

choose a variety that is resistant to major diseases. In

the garden, squash and other cucurbits should not be

planted in the same area and/or next to each other year

after year. Diseases that affect squash also affect other

cucurbits such as gourds, melons, cucumbers, oranges

and winter squash. If possible, allow pumpkins to

grow for at least 3 to 4 years to reduce diseases such

as Fusarium fruit rot, white spot, phytophthora and

white Mold. The longer the head and other cucurbits

are in the soil, the less likely soil diseases will occur.

Be sure to plant pumpkins in well-drained soil.

Waterlogged soil is a good environment for diseases

such as Phytophthora to grow. Mulching the soil with

straw, hay or fallen leaves to a depth of 6 inches will

help protect the fruit from contact with the soil and

will help reduce soil-borne diseases such as

Phytophthora and Fusarium. Mulching can also help

reduce weeds and retain moisture throughout the

growing season. Large enough for good weather.

Indoor plants create a microclimate in the shade that

is conducive to bacterial growth. Avoid overwatering

at all costs. Overwatering can inhibit bacterial growth

and help spread disease. Use a watering can or

sprinkler system to water your plants as much as

possible. This will keep the leaves dry. If you must

water, be sure to do so in the morning to allow

adequate drying during the day. Growing fruit in the

Efﬁcient Detection of Cucurbit Pepo Leaf Diseases Using Advanced Image Processing Techniques

471

garden during the warm season in early fall will cause

sunburn, especially if the field has lost leaves to leaf

disease. In addition to late-season diseases such as

Fusarium fruit rot and white Mold, harsh summers

can affect fruit ripening in the garden. Prevent and

control diseases such as powdery mildew, downy

mildew, anthracnose and vitiligo. For the above

diseases, use an insecticide containing a chemical

called chlorothalonil, which is often sold at local

garden centres and supermarkets. Copper fixative is

another medication that can be used to help control

conditions such as angular plaque. To control blight,

cucumber striped and spotted cucumber beetles

should be controlled early in the growing season

when seedlings emerge. Use insecticides or powdered

insecticides. Always read and follow the label when

using pesticides. New plants can also be covered with

a mat to prevent insects, aphids etc. from feeding on

the seeds. (Trapman, and, Jansonius, 2024)

Vibhor Kumar Vishnoi et al. introduced a project

that aimed to identify diseases in apple leaves using

advanced learning methods. They tackled problems in

detecting diseases in agriculture by using

Convolutional Neural Networks (CNNs) to recognize

diseases from pictures of leaves. They also used

methods like moving, tilting, resizing, zooming, and

flipping images to make their training data better,

which led to more accurate disease classification.

Khalid M. Hosny et al. presented a study which

aimed to improve the accuracy of detecting diseases

in plant leaves. They tackled problems in identifying

agricultural diseases by creating a new, simplified

deep CNN model that identifies complex hidden

features. This model combines with traditional Local

Binary Pattern (LBP) features, which help

understanding the texture details from images of plant

leaves.

Yafeng Zhao and colleagues presented a study

which aimed to tackle the issue of uneven datasets in

detecting plant diseases. They suggested a technique

that employs DoubleGAN, a type of double

generative adversarial network, to create high-

resolution pictures of diseased plant leaves, thus

evening out the dataset. The DoubleGAN method has

two steps: first, healthy and diseased leaves are fed

into a Wasserstein generative adversarial network

(WGAN) to produce 64x64 pixel images of diseased

leaves; second, a super-resolution generative

adversarial network (SRGAN) improves these

images to 256x256 pixels.

Siwar Bengamra and colleagues presented a study

that aimed improve the clarity and understanding of

deep learning models used in agricultural research.

Although deep learning has been very successful in

identifying and categorizing diseases in plants from

images of their leaves, these models often operate like

"black boxes," which means it's hard to see how they

make their decisions. To tackle this issue, the authors

suggested a new method in Explainable Artificial

Intelligence (XAI) called the saliency method, which

helps to explain the predictions made by models that

detect diseases in potato leaves. Their method

involves making certain changes based on

intermediate results from object detection to show

which parts of the input image are most important for

the model's predictions. They tested the effectiveness

of their method through both visual and numerical

experiments on models that detect diseases in potato

leaves using the PlantDoc dataset.

Yash Dusane and his team created a project which

is about automatically finding and naming diseases in

hibiscus leaves. Since hibiscus has important health

benefits in Ayurveda, it's very important to find leaf

diseases early. The team used special computer

methods to look at pictures of leaves and figure out

which ones were sick. They first used a special

grouping method called k-means clustering to spot

the sick leaves, then they took out important details

from the pictures. After that, they used a special way

of sorting called KNN (K-Nearest Neighbors) to

name the diseases in the hibiscus leaves. This new

way makes it better at finding and naming diseases in

hibiscus leaves.

3 PROPOSED WORK

Our Cucurbita pepo leaf disease detection is based on

the concepts of YOLOv7 and CNN. There are four

types of diseases, including pox, mildew, mosaic, and

leaf disease, as well as health diseases. With proper

training using a carefully constructed database of

2,000 high-resolution images, the model can detect

symptoms with high accuracy.

This method uses the best imaging techniques to

increase the quality and accuracy of the model. The

integration of YOLOv7 ensures the operation of the

system by making it possible to see the place of care

and determine the reality of the disease. In addition,

the CNN architecture is tuned according to different

diseases and events to obtain the best image quality.

3.1 Flow Chart

This flowchart outlines the different steps of the

Cucurbita Pepo Leaf Disease Detection System: it

starts with high-resolution leaf image collection and

preprocessing. The advanced image processing is

INCOFT 2025 - International Conference on Futuristic Technology

472

then followed by state-of-the-art deep learning

models, including YOLOv7 and Convolutional

Neural Networks (CNNs), to analyze the images in

successive stages of processing.

It uses a number of convolutional layers for

processing the images in differentiating healthy from

diseased leaves, extending further to locating the

specific diseases. The last layer in this model then

boxes the detected area and labels each one with the

category it falls into, such as Healthy, Downy

Mildew, Powdery Mildew, Mosaic Disease, or

Bacterial Leaf Spot. With both computational

efficiency and detection accuracy being equally

important, a solution that used CNN for feature

extraction and then used YOLOv7 for object

detection was proved optimal. Models like ResNet or

VGG16 or even MobileNet were taking much more

time to process a high-resolution image and

producing lower accuracy.

Figure 1: Flow Chart

3.2 Dataset

The system is based on a dataset of 2000 high-

resolution pumpkin leaf images, carefully selected to

train and test different deep learning models in the

System. The dataset consists of images of healthy

leaves and those infected by four common diseases,

namely Downy Mildew, Powdery Mildew, Mosaic

Disease, and Bacterial Leaf Spot. All classes within

this dataset have been 'balanced' for the model to

present diversity while training the data.

Table 1: Pumpkin leaf diseases dataset summary for

YOLO training.

Class Number of Samples

Health

400

Down

mildew 400

Powder

mildew 400

Mosaic diseases 400

Bacterial leaf spot 400

3.3

Modular Architecture (YOLOv7)

YOLOv7 is the chosen model because it is among the

most recent object detection systems and has

incomparable speed and accuracy in detecting and

locating certain objects in pictures.

Thus, with YOLOv7, the identification and

classification of diseases on Cucurbita pepo leaves

would be done perfectly. It is very fast in processing

high-resolution images with high precision, spotting

defect areas on the surface area of a leaf. It uses a

number of convolutional layers for processing the

images in differentiating healthy from diseased

leaves, extending further to locating the specific

diseases. The last layer in this model then boxes the

detected area and labels each one with the category it

falls into, such as Healthy, Downy Mildew, Powdery

Mildew, Mosaic Disease, or Bacterial Leaf Spot.

3.4 Training Setup

In this training setup, transfer learning was employed

by first pre-training the YOLOv7 model on a large

amount of image data and then fine-tuning it on

Cucurbita pepo leaf disease data. Configuration

entailed setting up loss functions, optimizers,

adjusting learning rates, and batch sizes among other

hyperparameters for optimal accuracy in leaf disease

detection and classification.

3.5

Disease Detection Module

The module involves the initialization and

management of a disease detection engine that uses

YOLOv7 in real-time object detection and CNNs for

further classification. It processes input images of the

Cucurbita pepo leaves for efficient detection and

classification of diseases like Downy Mildew,

Powdery Mildew, Mosaic Disease, and Bacterial Leaf

Spot. This module plays a very critical role in the

Efﬁcient Detection of Cucurbit Pepo Leaf Diseases Using Advanced Image Processing Techniques

473

generation of diagnosis results, which, in agriculture,

are of utmost importance in handling diseases

effectively.

3.6 Task Module

The Task Module defines the following actions to

perform by the system: identify diseases from

Cucurbita pepo leaves through input images. It

coordinates the entire process of image preprocessing

right up to the running of the YOLOv7 model,

performing the identification of diseases. Specific

arrangements will be made for the integration of

results so that the end users get diagnostic

information with suggestions on the management of

diseases.

3.7 Agent Module

The Agent Module assigns different agents to

different roles and objectives while aiming to enrich

Garden Management. The various agents include the

'Disease Diagnosis Agent', which, given an input

image, analyzes and classifies Cucurbita pepo leaf

diseases into disease categories; the 'Treatment

Advisor Agent' gives certain management strategies

or recommendations regarding the handling of the

detected disease. Each of these agents bringing

various aspects of disease detection and management

to the attention of users, giving supportive care that is

necessary for addressing gardening needs.

4 SIMULATION PARAMETERS

In this work, the dataset consisted of 2000 high-

resolution images of leaves of Cucurbita pepo, each

from five classes: 400 samples were labeled as

"Healthy," 400 samples as "Downy Mildew," 400

samples as "Powdery Mildew," 400 samples as

"Mosaic Disease," and 400 samples as "Bacterial

Leaf Spot." It is divided into training set and

validation set; 80% of its images will be used to train

and the remaining 20% to validate and test it in order

to guarantee generalization on new data. These are

high-resolution images that contain all the fine

details. Also, some of the preprocessing steps involve

resizing, normalization, and data augmentation.

YOLOv7 is set for real-time detection, while CNN

will be fine-tuned using optimal hyperparameters

relating to learning rate, batch size, and number of

epochs. These simulation parameters are relevant for

assessing model performance in the classification of

Cucurbita pepo leaves by an input image that should

be fed into it.

5 RESULT AND DISCUSSION

In this section, we report research results on cucurbit

leaf disease detection based on YOLOv7 and CNN

architecture. In the training process, we used 2000

high-resolution images of the five categories:

Healthy, Downy Mildew, Powdery Mildew, Mosaic

Disease, and Bacterial Leaf Spot. Therefore, we split

the data into two equal parts: the training subset

(80%) to train the model and the testing subset (20%)

to evaluate the model's ability. Below, we explain the

detailed performance analysis of the model, the

metrics used for the evaluation, and other techniques

for comparison purposes.

5.1 Model Performance

Criteria such as accuracy, precision, recall, F1 score,

and Competition Over Unity (IoU) were used to

describe YOLOv7 as a model. In fact, the model was

able to identify four types of diseases along with the

healthy leaves with higher accuracy during testing.

Notably, with YOLOv7's fast object detection

capability coupled with CNN's deep learning feature

set, it managed to directly localize and classify

disease spots quite precisely.

1.Accuracy: The overall accuracy of classification

on the test set was 96.5%. This large value ensures the

reliability of the disease symptoms detected by the

model on Cucurbita pepo leaves.

2. Precision and Recall: Table II shows that the

precision and recall values are high for each class. It

implies that the model can recognize both diseased

and healthy leaves without much false positive or

false negatives.

Table 2: Accuracy Measures Table.

Class Precision

(%)

Recall

(%)

F1-Score (%)

Health

97.2 96.5 96.8

Downy

mildew

96.0 95.7 95.8

Powdery

mildew

96.8 96.0 96.4

Mosaic

diseases

95.9 95.2 95.5

Bacterial

leaf s

95.4 95.0 95.2

INCOFT 2025 - International Conference on Futuristic Technology

474

3. IoU: The average IoU value of the predicted

and ground truth bounding boxes for the region was

attained as 91.8%, which implies that YOLOv7

correctly localizes the diseased region.

5.2

Comparison with other models

We compare the performance of the YOLOv7-CNN

model with previous works that include the

underlying ResNet, VGG16, and MobileNet models.

The results showed that YOLOv7 proved to be faster

compared to the mentioned models and produced

much higher accuracy.

5.3

Discussion

Results: The results validate the ability of the

proposed model in classifying with precision diseases

on the leaves of Cucurbita pepo. The success of the

YOLOv7 model in real-time detection opens new

avenues for integrating such technologies into

precision agriculture. Their high accuracy and low

false detection rates guarantee that the model will be

trusted by farmers and agronomists to be of use for

early diagnoses, thereby allowing appropriate

interventions at the correct time.

Thus, this research demonstrates the potential use

of AI-derived technologies for enhancing

productivity in agriculture through the reduction of

disease impairment on crops. Future work can focus

on incorporating a more comprehensive set of

conditions and then optimizing the model for an edge

device platform in monitoring in situ diseases.

Figure 2: Model Evaluation

Figure 3: Result

6 CONCLUSIONS

This study presents a deep learning method for the

detection and classification of cucurbit plant leaf

diseases using YOLOv7 and CNN. The model was

able to identify four diseases as healthy leaves with

96.5% overall accuracy from high-resolution images,

thus providing important information on monitoring

performance to improve agricultural disease

management practices. The application of AI

technology such as YOLOv7 will pave the way for

scalable systems that will improve crop disease

management for farmers moving forward.

REFERENCES

Bonkra, A., Pathak, S., Kaur, A., & Shah, M. A. (2024).

Exploring the trend of recognizing apple leaf disease

detection through machine learning. Artificial

Intelligence Review, 57, Article no. 21, January.

Khan, A. I., Quadri, S. M. K., Banday, S., & Shah, J. L.

(2022). Deep diagnosis: A real-time apple leaf disease

detection system based on deep learning. Computers

and Electronics in Agriculture, 198, Article no. 107093,

July.

Galbraith, S. (2024). Fresh Apples, Grapes, and Pears:

World Markets and Trade. World Production, Markets,

and Trade Report, June 18.

Vishnoi, V. K., Kumar, K., & Kumar, B. (2021). Plant

disease detection using computational intelligence and

image processing. Journal of Plant Diseases Protection,

128(1), 19–53, February.

Salcedo, A. F., Purayannur, S., Standish, J. R., Miles, T.,

Thiessen, L., & Quesada-Ocampo, L. M. (2021).

Fantastic Downy Mildew Pathogens and How to Find

Them: Advances in detection and diagnostics. Plants

(Basel, 10(3), 435, March.

Landschoot, P., Abbey, T., & Delvalle, T. (2024). Turfgrass

Diseases: Powdery Mildew (Causal Fungus: Blumeria

Graminis - Formerly Erysiphe Graminis). The

Pennsylvania State University, ART-7112.

Vinje, E. (2024). Mosaic Virus. Planet Natural. Available

at: https://www.planetnatural.com/pest-problem-

solver/plant-diseases/mosaic-virus/ [Accessed 15

September 2024].

Jarial, K., Jarial, R. S., Bhardwaj, P., Sharma, A., & Hallan,

S. (2023). Bacterial leaf spot of cucurbits: A menace to

cultivation. IntechOpen, Article no. 1001862, July 4.

Holb, I. J. (2009). Fungal disease management in

environmentally friendly apple production – A review.

(Vol. 2, pp. 219–292).

Trapman, M., & Jansonius, P. J. (2024). Disease

management in organic apple orchards is more than

applying the right product at the correct time. Organic

Agriculture, 5, 134–148.

Efﬁcient Detection of Cucurbit Pepo Leaf Diseases Using Advanced Image Processing Techniques

475