Experimental Evaluation of Deep Learning Based Plant Leaf Disease

Detection System Using Computer Assisted Image Processing

Techniques

P. Ramya, Mohanraj S., Mageshwaran N., Mohamed Shafeeq J. and Monishwar D. K.

Department of Computer Science and Engineering, Mahendra Engineering College, Tamil Nadu, India

Keywords: Deep Learning, Leaf Disease, Plant Disease Detection, Image Processing, Neural Classification Network,

ENCN, Support Vector Machine, SVM.

Abstract: One of the most important contemporary agricultural techniques, plant disease detection aids in the early

diagnosis of crop illnesses, which allows for more effective management and the prevention of substantial

losses. Color changes, spots, lesions, or structural malformations are common visible indicators of plants

damaged by diseases. Handheld, drone-mounted, or integrated into smart agricultural automation systems,

high-resolution cameras or sensors record these symptoms. Computer Vision and Deep Learning (ML)

algorithms examine the gathered data for patterns in form, texture, and color to determine the presence of

illnesses. To achieve accurate disease identification in plant leaves, this paper proposes a novel deep learning

model which is Enhanced Neural Classification Network (ENCN). So, the performance of the model can be

tested by cross-validating it with a conventional learning scheme Support Vector Machine (SVM). An

accurate diagnosis enables farmers to implement timely countermeasures against diseases such as blast,

bacterial blight or powdery mildew. The system in many cases will recommend the use of pesticides, changes

to that amount of water or fertilizer that is applied to crops or quarantining sick plants to stop disease from

spreading, she said. The technology improves productivity, ensures accuracy, reduces costs, and promotes

sustainable farming practices by utilizing the resources that are already there more effectively. Recent

advancements in plant disease detection including integration of the internet of things and drone monitoring

result into crop management, high yield and sustainable agriculture environment. Beyond aiding early

diagnosis and management, predictive analysis from plant disease detection systems, based on patterns in

historical and environmental data, enables farmers to prepare for future crop disease outbreaks. For all, these

systems make it possible to monitor a vast area, which saves time and effort when evaluating the health of

enormous farmlands.

1 1 INTRODUCTION

Plant disease detection is an essentially significant

field of study using Deep Learning for the detection

and diagnosis of plant diseases (Wubetu Barud

Demilie,2024). Since they may seriously affect

agricultural production, it is very important to

diagnose these diseases in time and to take

precautions against them to guarantee food security

and healthy crops. (Payal Trivedi, et al.,2024) Firstly,

the process would have required a lot of time and

money from the specialists who had to work out, on a

laborious and error-prone way, the identification of

plant diseases. The following figure, Figure 1

represents the dataset image samples.

However, with the advent of AI and ML, one can

now automate the identification of plant diseases with

high accuracy and speed in modern agriculture, which

is very promising (3. Manjunatha Shetti gere Krishna,

et al.,2025).

One of the common techniques that many

researchers employ is using supervised learning to

build machine learning models for plant disease

diagnosis. The algorithms are trained over labeled

datasets, wherein the plant images are divided into

either healthy or diseased types of images (Priyanka

Pradhan, et al., 2024). The model learns to discriminate

between healthy and diseased plants by analysis of

millions of images and considered variables such as

color, texture, shape and some presence of unique

patterns characteristic to several diseases.

Ramya, P., S., M., N., M., J., M. and K., M.

Experimental Evaluation of Deep Learning Based Plant Leaf Disease Detection System Using Computer Assisted Image Processing Techniques.

DOI: 10.5220/0013867400004919

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

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 1, pages

443-452

ISBN: 978-989-758-777-1

443

Figure 1: Dataset Image Samples.

The advantage of proposed model ENCN lies in

their ability to process complex visual information

with the built-in automatic extraction of relevant

features from images. That is exactly why they are

used to mark subtler signs of plant diseases in

situations where conventional methods might not

work very well. The performance of the detection

system relies chiefly upon the quality of the dataset

on which the machine-learning models are trained.

The datasets contain both healthy and diseased plants,

including quality images of plants grown under

managed conditions. Datasets also have labels to

indicate the presence or absence of diseases. Several

of such datasets contain examples of three highly

frequent agricultural diseases: leaf rust, powdery

mildew, and blight; in training models to detect signs

of disease, these diseases are of outstanding

importance in biological and agricultural research.

(B.V. Nikith, et al., 2023). A robust model would

require a large, diverse and balanced dataset, which is

not very easy to establish (Cemal Ihsan SOFUOGLU, et

al., 2024). This dataset should include a range of plant

species and diseases.

As research progresses and various datasets

become more widely available, this will become a

useful tool for farmers to safeguard crops and,

therefore, increase global food security by providing

enhancements in the speed, accuracy, and scalability

of machine-learning algorithms to identify instances

of plant diseases (Alwan Fauzi, et al.,2023). First, one

has to collect a large number of datasets containing

images of good and poor conditions of plants as a

step-in kernel research of machine learning for plant

disease detection. Images of plants collected to be

labeled to show the absence or presence of certain

diseases have been captured by devices like mobile

phones, drones, or digital cameras. Next, these

images undergo preprocessing to enhanced form

features for appearance. This includes resizing,

normalization, and augmentation (such as rotating,

flipping, or scaling) in order to achieve data

heterogeneity, thus preventing overfitting. Image

preprocessing permits the machine learning model to

concentrate on the main features such as shape, color,

and texture, which are critical in the detection of

disease (Rashmi Ashtagi, et al.,2025). An important

aspect of autonomous feature extraction is deep

learning models such as CNNs, which enable the

system to internally learn complex patterns of input

images (Ashutosh Kumar Singh, et al.,2022). After

training on a labeled dataset, a model's performance

is evaluated using various metrics such as accuracy,

precision, and recall. In this case, a validation set is

used for this purpose. The model is capable of being

adjusted as necessary, in order to improve its

performance. Once the model is accurate enough, it

will be used in real-time applications, just like online

platforms or mobile applications, so that farmers can

identify disharmony in his/her crops by uploading

pictures. In order to help guide farmers toward the

rapid detection of possible illnesses, the software

analyses these images and provides predictions. In

addition, the system may improve itself over time in

terms of different plant species and climatic

conditions, since it learns from new data. There is

great potential for machine-learning-powered plant

disease detection to revolutionize the existing

agricultural sector by reducing losses on crops and

increasing food security.

The process of automating and improving the

accurate and timely diagnosis of plant disease can be

achieved through machine learning, especially via

convolution neural networks. Convolutional neural

networks, unlike any other machine learning

algorithm, learn salient features from raw image data,

such as pattern, texture, and shape specific to the plant

disease. Training on a large dataset of labeled images

allows convolutional neural networks to detect some

early signs of diseases that might not be perceivable

by the human eye. This allows prompt diagnosis,

enabling farmers to take corrective or preventive

measures before the disease spreads and causes large

crop losses (Kamaldeep Kaur, et al.,2024).

Wzw the ultimate goal is to implement an easy-

to-use living system to allow real-time diseases

identification in crops using different platforms such

as mobile application; this will allow farmers better

management of crop health and hence increased

agricultural yield.

(i) To enrich the model and avoid overfitting, image

synthesis is used to create additional images of

infested plants and hence augment the dataset.

(ii) In general, such approaches produce artificial

images of sick plants by training their convolutional

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

444

neural networks with the data definitions provided

through GANs. Where there is the absence of

adequate labeled datasets, this can prove useful.

(iii) Early prediction and curing of plant diseases and

minimizing losses to crops are made easier through

the use of synthetic images produced by image

synthesis, which increases the accuracy of disease

detection models.

(iv) Synthetic images of infected plants can be used

to fine-tune pre-trained CNN models. This allows

efficient transfer learning and more accurate disease

diagnosis.

2 RELATED WORKS

Plants form the basis of the world's food supply;

nonetheless, plant diseases cause considerable losses

in the output of crops, related to many environmental

conditions (Muhammad Shoaib, et al.,2025).

Nevertheless, Tropical plant disease identification

undertaken by people becomes a long and tedious

task. It is not always very reliable as a tool in Plant

Disease detection and control. One way to tackle

these difficulties is by implementing modern

technologies like Deep Learning (DL) and Machine

Learning (ML). These will allow for the early

detection of plant illnesses. This study delves into the

latest developments in plant disease diagnosis using

ML and DL approaches. The trials included in this

paper show that these methods can improve the

efficiency and accuracy of plant disease detection,

and the research focuses on publications from 2015 to

2022. Besides plant disease recognition, this paper

also covers the challenges and constraints of ML and

DL for plant diseases, such as lack of data, poor

images, healthy versus sick plant distinction, etc. The

survey provides a comprehensive review of the state-

of-the-art work on the detection of plant diseases,

along with their pros and cons, and recommendations

to overcome the challenges faced while employing

them. As a result, it is valuable for researchers,

practitioners, and industry professionals in this field.

Scientists are now adopting intelligent

agriculture-a means of greatly enhancing production

with the incorporation of the use of Artificial

Intelligence (AI)-to deal with various problems

within agricultural fields (Sherihan Aboelenin, et

al.,2025). There are a lot of illnesses that harm crop

yields, and there are a lot of plants in the globe, thus

finding and classifying plant diseases isn't an easy

task. Any AI-based system aims to accurately

classify plant illnesses and detect them early. In order

to greatly enhance the accuracy of plant leaf disease

categorization, this research suggests a hybrid

architecture. The power of Convolutional Neural

Networks (CNNs) and Vision Transformers (ViT) is

utilized in this suggested model. Robust global

features are extracted using an ensemble model that

includes the popular CNN designs VGG16,

Inception-V3, and DenseNet20. The next step in plant

disease detection with high accuracy is applying a

ViT model for local feature extraction. Under testing

in the apple and corn public datasets, there are four

classes per each dataset. The apple dataset has an

accuracy of 99.24% while the corn dataset has 98%.

This hybrid model will efficiently ascertain and

classify multi-class plant leaf diseases in reference to

other similar published models.

To protect agricultural crop output and guarantee

food security, early and precise identification of plant

leaf diseases is of the utmost importance (Sasikala

Vallabhajosyula, et al.,2024). Bacteria, fungus,

weather, and other environmental variables are

among the many causes of leaf diseases that plants

experience during their life cycles. By combining the

best features of the enhanced Vision Transformer

with ResNet9, the authors of this study provide a new

hierarchical residual vision transformer that can help

with the early diagnosis of leaf illnesses. By lowering

the number of trainable parameters and using fewer

calculations, the suggested model is able to extract

more relevant and discriminating features. Tests

using 13, 38, and 51 distinct leaf disease classes are

conducted on the Local Crop dataset, the Plant

Village dataset, and the Extended Plant Village

Dataset, respectively, to assess the efficacy of the

suggested approach. Using ResNet 9 for feature

classification and the optimal trail parameters from

Improved Vision Transformer, the suggested model

is trained. When tested on the aforementioned

datasets, the suggested model beat competitors like

InceptionV3, MobileNetV2, and ResNet50 across a

variety of metrics.

Agriculture is an essential need and their primary

source of domestic income for many countries (Anuja

Bhargava, et al.,2024). Plant diseases effected by

more than one pathogen (as in bacteria, fungus and

viruses) are so common that agricultural corporations

lose big bucks worldwide. It is critical to monitor

plant diseases in order to ensure the quantity and

quality of harvests. This highlights the critical nature

of plant disease detection. Symptoms of the plant

disease syndrome manifest in certain plant tissues.

Still, individual plant leaves are usually the first to

show signs of infection. Several researchers have

used computer vision, deep learning, few-shot

learning, and soft computing approaches to

Experimental Evaluation of Deep Learning Based Plant Leaf Disease Detection System Using Computer Assisted Image Processing

Techniques

445

automatically detect plant diseases from leaf images.

Quick and suitable efforts to prevent a decrease in

crop quality and quantity can also be achieved by

farmers using these strategies. By using these

methods to illness recognition, we may speed up

technology and research while avoiding the

drawbacks of origin by avoiding factious feature

selection and extraction. Additionally, specific

molecular methods have been developed to forestall

or lessen the impact of the infectious danger. Thus,

this research assists the researcher in designing

automated plant disease identification systems using

deep learning, machine learning, and few shots of

learning. It also gives specific diagnostic methods to

prevent disease. We also discuss some of the next

steps in illness categorization.

Reducing economic repercussions and optimizing

agricultural output require precise and timely

detection of plant leaf diseases (Eman Abdullah

Aldakheel, et al.,2024). The problem with precisely

identifying certain illnesses is that farmers rely on

traditional manual approaches, which makes it

difficult. Applying the YOLOv4 algorithm to the

problem of plant leaf disease detection and

identification is the focus of this study. The big Plant

Village Dataset is composed of more than 50,000

pictures of healthy and diseased plant leaves from

fourteen different species; the study prepares the very

general advanced systems for prediction of

agricultural diseases. To enhance the dataset and

fortify the model's durability, data augmentation

techniques such horizontal flip and histogram

equalization were employed. We compared the

YOLOv4 algorithm's performance against that of

other well-known target recognition methods, such as

Densenet, Alexanet, and neural networks, as part of

our thorough evaluation. An astounding 99.99%

accuracy was attained by YOLOv4 when applied to

the Plant Village dataset. The proposed method was

verified based on consistently very high values, with

0.99 scores for all the following metrics: accuracy,

precision, recall, and F1-score. The findings of this

study highlight the remarkable progress made in plant

disease detection and highlight the potential of

YOLOv4 as an advanced tool for precise disease

prediction. Because they increase our ability to

manage diseases and safeguard crops, these

innovations are incredibly important for researchers,

farmers, and everyone else working in the agricultural

sector. After the model is developed, it has to be

trained using a large number of pictures that have the

necessary objects tagged. Keep in mind that the

EfficientNetV2B1model can only learn to

differentiate between the classes if the input is

balanced. After that, you need to feed the data into

the EfficientNetV2B1 model. Dataset size

determines whether this is best done in batches or in

one continuous run. Next, a suitable optimizer, such

as Adam or SGD, has to be used to train the model.

In order for the model to learn to identify different

objects in the images, its weights are modified

continually during the training phase. After then, the

testing set may be used to assess the model's

correctness. The model's performance may be

evaluated using a variety of measures, including

recall, precision, and F1 score. We may measure the

model's performance by keeping track of how many

images are appropriately classified.

3 METHODOLOGY

Deep learning for the recognition of diseases in plants

employs deep learning and some old techniques. In

general, visual-based plant disease classification has

leveraged conventional deep learning techniques,

such as Random Forest, Naive Bayes, Support Vector

Machines, and K-Nearest Neighbors. On the other

hand, deep learning such as convolutional neural

networks has exhibited exceedingly effective

performance in identifying plant diseases present in

photos. Transfer learning employing pre-trained

ENCN models is yet another application for the

identification of plant diseases. Hybrid techniques,

combining methods such as SVM with ENCN, have

also been developed to capture the best of both

worlds. Other methods that help prepare the dataset

and enhance it are data augmentation, image

processing, and feature extraction techniques. Upon

using popular datasets such as Plant Village and IPM

Images for training and testing these models, deep

learning shows a promising technique for reliable and

quick plant disease identification. Here are some

drawbacks of the procedure.

• Inadequate or biased training data might

cause plant disease detection algorithms to

incorrectly diagnose illnesses.

• Models for plant disease detection based on

deep learning are difficult to implement on

low-power devices due to the high memory

and processing requirements of these

models.

• High-quality images are crucial to

conventional plant disease detection models,

but these images are susceptible to errors

caused by things like lighting, camera

quality, and image processing.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

446

The first phase of collecting data for the system

would involve collecting a large dataset of healthy

and diseased plants images from different sources.

After this, images should be normalized to a certain

size, their pixel values normalized, and data should be

enhanced and diversified through data augmentation

techniques for use in analysis. To be able to

accomplish this, a ENCN that consists of

convolutional layers, pooling layers, and fully

connected layers is trained. By using k-fold cross-

validation, taking precision, recall, and F1 score

along with accuracy matrixes for the evaluation

comes the next part on the addition of images from

the users for diagnosis of the specific plant species or

diseases via transfer learning which can allow fine-

tuning. Certainly, asked for diagnosis, this web or

mobile application is also usable for resolvable fine-

tuning through transfer learning for a specific plant

species or diseases. Indeed, easy usage interface in

the system provides sufficient support for specialists

in agriculture and farmers to utilize it for disease

diagnosis and treatment suggestions. The system,

therefore, can combine and work together with

various technologies, like satellite imaging or drones,

allowing small and large-scale infection detection and

monitoring. The following figures, Figure 2 and

Figure 3 show the flow diagram and system

architecture of the proposed approach.

Figure 2: System Flow Diagram.

Figure 3: System Architecture.

The process of plant disease detection starts with

acquiring a large dataset of images from healthy and

diseased plants and working on them to standardize

diversity. This is followed by designing a ENCN

architecture and training using the preprocessed

dataset, resorting to methods like data augmentation

and transfer learning to facilitate better learning.

Subsequent to this step, accuracy and F1-score

metrics should be used to assess the model and deploy

it through a web or mobile application so that farmers

and agricultural specialists can upload images for

disease detection. Eventually the model continues to

get new updates from data again to learn and improve

its performance in providing proper diagnosis with

the required treatment suggestions. Below are the

advantages:

• The use of deep learning in the ENCN

algorithms for plant disease recognition can

have several benefits; it, first and foremost,

removes the long, tedious, and error-prone

human examination and diagnosis of plant

diseases. Such detection, therefore, becomes

more efficient and more accurate.

• ENCN models train actually to recognize

diseases early by noticing features and

patterns within the images that are difficult

for a human to perceive. In addition, deep

learning systems can learn through huge

datasets and continuously improve their

performances in detecting some diseases.

Experimental Evaluation of Deep Learning Based Plant Leaf Disease Detection System Using Computer Assisted Image Processing

Techniques

447

• With precise insights and treatment

recommendations from the ENCN algorithm

that can be performed by the agricultural

experts and farmers, preemptive measures

will, thus, be taken to control the outbreak of

the diseases.

3.1 Data Collection

It is challenging to guarantee the efficacy of transfer

learning because the initial MobileNet pre-trained

model was trained on the ImageNet dataset, which

does not only include the pictures that are required.

So, in order to train the model, we require a dataset

that includes trash photos. There is currently no fixed

dataset that is universally used for trash

categorization jobs. The TrashNet dataset is used for

rubbish classification; however it doesn't adequately

reflect the real situation of residential waste

categorization in India since it contains too few

categories. Consequently, this article builds a dataset

specifically for visual trash sorting using network

retrieval and real-world scene imaging in the lab,

covering both single-object and multiple-scene

scenarios.

3.2 Image Processing

To augment and normalize images of plants so that

accuracy in identification of plant disease is

improved, an Image Preprocessing obtains images.

Depending on the kind of input image it is processing,

it resizes images into fixed resolution, normalizes

pixel values, incorporates some augmentation

techniques, removes noise or other irrelevant

qualities, and enhances image quality with

brightening, contrast, and saturation enhancing. The

whole model is thus reinforced, has less influence

from external variables, and improves the accuracy in

diagnosis. The current discussion revolves around

whether a new medium can replace or precedes

existing media, particularly when the new channel

offers similar functionality as the old one. To perceive

the impact of a novel media on a preexisting one, it is

crucial to analyze customer perspectives about the

Modern channel and its potential to substitute the

current one.

3.3 User Interface

Dataset was divided into three subsets: validation,

training, and testing. User Explanation provides a

user-friendly interface for farmers and agricultural

experts to receive disease diagnosis and

recommendations for treatment. It explains the

detected disease, its symptoms, causes, and

prevention methods, enabling users to take informed

decisions about crop management and disease

control. Also, the process will provide users with

personal advice on pesticide application, fertilizer

application, and irrigation management, allowing

users to optimize their farm management systems. It

also provides information on how the disease may

potentially affect crop yield and quality, enabling

users to plan accordingly. Likewise, a feature on the

process will track the disease's history and give alerts

for possible outbreaks of the disease, allowing users

to implement some of their proactive measures to

avoid the spread of any disease.

3.4 ENCN Architecture

Following preprocessing, we select an ENCN

architecture that works best for trash picture

classification. Our classification model of choice is

EfficientNetV2B1. We loaded the model weights

from a source similar to EfficientNetV2B1. The

models can be found within deep learning tools,

namely, TensorFlow or PyTorch. The pretrained

EfficientNetV2B1 model has been modified by

removing the previous classification head (fully-

connected layers) and inserting a new classification

head corresponding to the number of garbage

categories in our dataset. We used ENCN layers as

feature extractors to educate the model in such a way

as to extract hierarchical and discriminative qualities

from the garbage images. It is thereby necessary to

use an optimizer, such as AdamorSGD, to work to

adjust model weights through back propagation, as

well as a suitable categorical cross-entropy loss

function for multi-class classification.

3.5 Model Training

Following model creation, it is necessary to train the

model using a large number of photographs annotated

with the necessary items. If you want the

EfficientNetV2B1model to learn to differentiate

between the different classes accurately, you must

ensure that the data is balanced. Once the data is

available, it has to be loaded into the

EfficientNetV2B1 model. Depending on the dataset

size, this can be done in batches or all at once. Then

we proceed to model training, using an appropriate

optimizer (Adam or SGD). During training, the model

adjusts its weights to recognize objects within the

images. In the next step, the model could be tested

with the testing set to check its performance. Recall,

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

448

accuracy, and F1 score are all some ways we can

evaluate the model. A common way to evaluate

model performance is by quantifying how many

photos it can correctly label. It can be used on unseen

data to further test the generalizability of the model.

This will demonstrate how well the model can

predict using unobserved data.

4 RESULTS AND DISCUSSION

Figure 4: Dataset Uploading Port.

Using deep learning techniques to plant leaf

disease detection systems has improved agricultural

diagnostics extensively. The proposed

representations are fed into different deep learning

architecture such as Enhanced Neural Classification

Network (ENCN) to identify and categorize the

diseases based on leaf images. Consequently, they

facilitate early intervention and improved crop

management. Applications of deep learning models

like the ENCN in Android apps make real-time

detection and classification of maize and other cereal

problems fast and efficient. Precision farming

undergoes a paradigm shift with the implementation

of deep learning for the identification of plant leaf

maladies. The high accuracy and efficacy of these

models suggest that they may be widely accepted in a

variety of agricultural settings.

Figure 5: Validation Image Uploading Port.

Future research should focus on the inclusion of

these systems in readily accessible platforms for

farmers and other experts, the enhancement of model

resilience under diverse environmental conditions,

and the expansion of the diversity of plant species and

illnesses that are addressed. Figure 5 clearly shows

the testing image uploading portal that was created

using an Android application; Figure 4 shows the

dataset uploading site of the suggested method. Image

pre-processing and leaf disease prediction results of

the suggested approach are shown in Figures 6 and 7,

respectively.

Figure 6: Image Pre-Processing.

Experimental Evaluation of Deep Learning Based Plant Leaf Disease Detection System Using Computer Assisted Image Processing

Techniques

449

Figure 7: Disease Prediction.

Figure 8 shows the results of a cross-validation

test between the suggested model ENCN and a

traditional learning model known as SVM, which was

used to determine the model's prediction accuracy.

Table-1 provides a descriptive representation of the

same.

Table 1: Prediction Accuracy Comparison between SVM

and ENCN.

Epochs SVM (%) ENCN (%)

100 88.62 96.27

125 86.61 97.63

150 84.43 95.87

200 85.71 97.31

250 86.62 97.11

300 87.62 97.35

350 87.63 97.58

400 88.62 97.82

450 89.73 96.79

500 88.92 97.29

Figure 8: Model Prediction Accuracy.

The proposed ENCN loss ratio assessment is

presented in the following figure, Figure 9. In this

assessment, we will cross-validate the previous

scheme's proposed scheme with a classical learning

model referred to as SVM to find the proposed

scheme's loss ratio. The same is described in the next

table called Table 2.

Table 1: Comparison of Loss Ratio Between Svm and Encn.

Epochs SVM ENCN

100 6.34 1.19

125 6.62 1.26

150 4.79 1.34

200 6.27 1.82

250 6.36 1.94

300 5.29 2.65

350 4.73 2.72

400 4.50 2.66

450 6.69 2.67

500 6.47 2.88

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

450

Figure 9: Loss Ratio.

5 CONCLUSIONS

Plant disease detection detected plant sickness using

ENCN deep learning and computer vision approach

have proven highly effective in accurately identifying

and tackling plant disease. By automating the process

of disease detection, this approach reduces manual

labor, enhances productivity, and allows for early

diagnosis of diseases, minimizing damage to crops

and losses. Moreover, the disease diagnosis accuracy

of the system is high, thereby minimizing the chances

of treating wrong disease. It has the potential to

develop further in the future through transfer

learning, more disease detection and IoT sensors

integration. In summary, the Plant Disease Detection

system is a powerful tool that aids farmers and

researchers alike in maintaining optimal crop health

and productivity, playing a crucial role in ensuring

global food security and sustainability. The Plant

Disease Detection system has far-reaching

implications for global food security and

sustainability. With unparalleled disease diagnosis

accuracy, it ensures that researchers and producers

have a reliable tool to prevent misdiagnosed and

mismanaged. Potential for integration with IoT

sensors, multi-disease detection, and transfer learning

provides a wide avenue for further enhancement of

this system which in turn will help for having more

sustainable and resilient agricultural systems.

In addition, the system can be adapted to detect

several illnesses at once, which decreases the need

for separate models and increases overall efficiency.

The application of transfer learning will also enable

the system to adjust to new, previously unseen

ailments while alleviating the need for significant

retraining. An app will make the system easy to use

which will promote the wide use of the system

between farmers and researchers. By incorporating

explain ability and interpretability methods, users will

be able to understand how and why the system came

to a particular decision, leading to increased trust and

understanding amongst users. Transfer learning: The

system must be able to adapt with the new unknown

diseases, without retaining from scratch. The

development of explainable/ interpretable techniques

will provide insights into how the system makes

decisions, while a future mobile application will act

as a bridge between farmers and researchers.

REFERENCES

Alwan Fauzi, et al., "Development of a Mobile Application

for Plant Disease Detection using Parameter

Optimization Method in Convolutional Neural

Networks Algorithm", International Journal of

Engineering Technology, 2023.

Anuja Bhargava, et al., "Plant Leaf Disease Detection,

Classification, and Diagnosis Using Computer Vision

and Artificial Intelligence: A Review", IEEE Access,

2024.

Ashutosh Kumar Singh, et al., "Hybrid Feature-Based

Disease Detection in Plant Leaf Using Convolutional

Neural Network, Bayesian Optimized SVM, and

Random Forest Classifier", Journal of Food Quality,

2022.

B.V. Nikith, et al., "Leaf Disease Detection and

Classification", Procedia Computer Science, 2023.

Cemal Ihsan SOFUOGLU, et al., "Potato Plant Leaf

Disease Detection Using Deep Learning Method",

Journal of Agricultural Sciences, 2024.

Eman Abdullah Aldakheel, et al., "Detection and

identification of plant leaf diseases using YOLOv4”

Frontiers in Plant Science, 2024.

Kamaldeep Kaur, et al., "Enhancing Plant Disease

Detection using Advanced Deep Learning Models",

Indian Journal of Science and Technology, 2024.

Manjunatha Shettigere Krishna, et al., "Plant Leaf Disease

Detection Using Deep Learning: A Multi-Dataset

Approach", J, 2025.

Muhammad Shoaib, et al., "An advanced deep learning

models-based plant disease detection: A review of

recent research Updated", Frontiers in Plant Science,

2023.

Experimental Evaluation of Deep Learning Based Plant Leaf Disease Detection System Using Computer Assisted Image Processing

Techniques

451

Payal Trivedi, et al., "Plant Leaf Disease Detection and

Classification Using Segmentation Encoder

Techniques", Open Agriculture, 2024.

Priyanka Pradhan, et al., "Plant disease detection using leaf

images and an involutional neural network",

Environment Conservation Journal, 2024.

Rashmi Ashtagi, et al., "Fusion of AI Techniques: A Hybrid

Approach for Precise Plant Leaf Disease

Classification", Journal of Electrical Systems, 2025.

Sasikala Vallabhajosyula, et al., "A novel hierarchical

framework for plant leaf disease detection using

residual vision transformer", Heliyon, 2024.

Sherihan Aboelenin, et al., "A hybrid Framework for plant

leaf disease detection and classification using

convolutional neural networks and vision transformer",

Complex Intell. Syst., 2025.

Wubetu Barud Demilie, "Plant disease detection and

classification techniques: a comparative study of the

performances", Journal of Big Data, 2024.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

452