Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr2394

Plant Disease Detection

N Bhavana1; P Likithasree2
1
Assistant Professor; 2Post Graduate
1;2
Department of MCA, Annamacharya Institute of Technology & Sciences, Tirupati, AP, India.

Publication Date: 2025/05/12

Abstract: Plant diseases pose a major danger to agricultural productivity and global food security. In order to automatically
detect plant diseases, this study presents a deep learning-based technique for categorising leaf photos. The system uses
Convolutional Neural Networks (CNNs) constructed in PyTorch to identify 39 different forms of plant diseases using the
PlantVillage dataset. A pre-trained model is integrated into an intuitive Flask web application, allowing users—farmers in
particular—to submit leaf photographs and receive prompt, accurate diagnoses. The model learns intricate visual patterns
associated with many plant diseases, offering an efficient, scalable, and cost-effective method for early disease diagnosis and
control in agriculture.

Keywords: Plant Health Monitoring, CNN Classification, Leaf Disease Detection, Smart Farming, Precision Agriculture.

How to Cite: N Bhavana; P Likithasree (2025) Plant Disease Detection. International Journal of Innovative Science and Research
Technology, 10(4), 3249-3252. https://doi.org/10.38124/ijisrt/25apr2394

I. INTRODUCTION of 39 distinct disease categories. The PlantVillage dataset, a

sizable, openly accessible collection of tagged plant leaf
Plant diseases have long been a major problem in photos that has established itself as a common benchmark in
agriculture, having a negative impact on food quality, plant disease classification research, is used to train the
agricultural output, and the general well-being of rural model. The CNN architecture is built and trained using
communities. Plant disease diagnosis has historically mainly PyTorch, an open-source machine learning framework.
depended on human skill, with pathologists' or farmers' visual Strong performance across a variety of real-world
inspection being crucial in identifying disease symptoms. circumstances is ensured by the trained model's ability to
However, particularly in rural and under-resourced areas, generalize successfully on unseen data.
these approaches are frequently constrained by human error,
inconsistent judgments, and the availability of expert The trained model is deployed via a Flask-based web
information. Furthermore, even professionals find it application, bridging the gap between research and practical
challenging to accurately discern between the visual signs of usage. Through this user interface, anyone—farmers and
many diseases since they are so similar. These difficulties agricultural professionals in particular—can upload photos of
highlight the necessity of an automated, precise, and afflicted plant leaves and get real-time forecasts about the
expandable method for diagnosing plant diseases. disease type. The system provides a useful tool for early
disease diagnosis, which is crucial for prompt intervention
Recent advances in artificial intelligence and deep and mitigation, by integrating machine learning into an
learning have expanded the possibilities available for approachable platform.
precision agriculture. Among these, image-based disease
diagnosis using Convolutional Neural Networks (CNNs) has Incorporating deep learning into agricultural diagnostics
demonstrated promise. Because CNNs are particularly adept improves disease detection accuracy while enabling resource-
at processing visual information, they can be used to constrained users to make well-informed choices. This
categorise plant diseases based on photos of leaves. By strategy therefore improves crop health management, lessens
learning to identify unique patterns and features in images reliance on chemical treatments, and makes a substantial
that may be hard for the human eye to notice, they increase contribution to sustainable agricultural methods. This study
the accuracy of diagnosis. shows how technology and agriculture may be combined to
successfully and efficiently address real-world problems.
This research aims to develop a deep learning model
capable of accurately classifying plant leaf pictures into one

IJISRT25APR2394 www.ijisrt.com 3249

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr2394
II. METHODOLOGY

 Dataset:

Fig 1 Records of Plant Illnesses

The project's dataset includes comprehensive records of III. SYSTEM IMPLEMENTATION

plant illnesses that impact crops including apples, corn,
grapes, and peaches, as well as their healthy counterparts. The suggested system for plant disease diagnosis is
Each entry includes the plant species, disease name, a implemented using deep learning algorithms and a web-based
descriptive summary of the symptoms, recommended interface. Initially, a sizable dataset of photos of both healthy
treatment steps, and a corresponding image URL. This and sick plant leaves was collected, and preprocessing
multimodal structure enables both visual and textual analysis. techniques like resizing, normalisation, and augmentation
The data was collected and curated from publicly available were used to boost the model's resilience. Next, the dataset's
agricultural resources and plant pathology databases. variability was increased by applying augmentation
techniques like flipping, rotation, and zooming. Using the
It ensures diversity in disease patterns, stages of PyTorch framework, a Convolutional Neural Network (CNN)
infection, and treatment approaches. The dataset supports was developed to categorise the plant diseases. ReLU
training and evaluating machine learning models for disease activation functions, pooling layers to lower dimensionality,
classification and decision-making. Preprocessing steps several convolutional layers for feature extraction, and fully
involved cleaning textual data and validating image links to connected layers to carry out the final classification make up
ensure consistency and accuracy. The overall dataset is the architecture. The top-performing model was saved as a.pt
structured to facilitate robust and real-time plant disease file for deployment following a rigorous training and
diagnosis systems. validation process. Flask was used to construct a web
application that would allow users to access the system. An
 Data Preprocessing image of a plant leaf can be uploaded to the program, which
The dataset went through several preparation then processes it, applies the trained model, and delivers the
procedures to ensure data quality and model readiness. To projected disease and further details. Structured CSV files
preserve data integrity, missing or null values were first found with thorough descriptions of every condition are the source
and eliminated. To enhance NLP-based analysis, text of this extra data. While the frontend provides an easy-to-use
descriptions were cleaned by deleting stop words, converting interface for user interaction, the backend manages image
to lowercase, and removing punctuation. Links that were processing, model loading, prediction, and information
broken or inaccessible were removed, and image URLs were retrieval. HTTP requests are used to control communication
verified. Treatment steps were lemmatized and tokenized for between the frontend and the backend. To facilitate hosting
uniformity. To make each illness entry compatible with on Local servers was supplied for deployment. A
machine learning algorithms, it was encoded into categorical requirements.txt file contained all necessary dependencies to
labels. To avoid bias toward more prevalent diseases, the guarantee a simple installation process.
dataset was also balanced. In order to guarantee uniformity
throughout the dataset, features were lastly standardized and
organized.

IJISRT25APR2394 www.ijisrt.com 3250

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr2394
IV. RESULTS AND ANALYSIS

Fig 2 Training and validation Accuracy

The graph illustrates the correlation between the number without experiencing excessive overfitting. The two curves
of training epochs and the accuracy the model achieved on do not abruptly diverge, indicating that the model maintains
the training and validation datasets. From 70% at the first its high capacity for generalisation to new data. The model
epoch to nearly 97% by the 25th epoch, the training accuracy gets increasingly better at properly classifying plant illnesses
shows a consistent rise with the number of epochs. In a as training goes on, as evidenced by the consistent
similar vein, the validation accuracy exhibits a closely improvement in validation accuracy throughout epochs. This
aligned trajectory, peaking at approximately 96% at the final encouraging pattern attests to the CNN architecture's
epoch after beginning at 68%. resilience as well as the efficiency of the preprocessing and
augmentation techniques used during model training.
The training and validation accuracy curves' parallel
growth shows that the model is learning efficiently and

Fig 3 Confusion Matrix Heatmap

IJISRT25APR2394 www.ijisrt.com 3251

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr2394
V. DISCUSSION REFERENCES

Creating an effective machine learning model for the [1]. Abbas, A., Zhang, Z., Zheng, H., Alami,
detection and classification of plant diseases was the aim of M.M., Alrefaei, A.F., Naqvi, S.A.H. et al.
this study. The model's classification accuracy was high, (2023) Drones in plant disease assessment, efficient
correctly predicting the majority of samples in each category, monitoring, and detection: a way forward to smart
as seen by the confusion matrix heatmap that visualised the agriculture. Agronomy, 13, 1524.
results. The high values that are highly concentrated along the [2]. Abdulridha, J., Ampatzidis, Y., Qureshi, J. & Roberts,
diagonal line of the matrix show that the model can P. (2022) Identification and classification of downy
distinguish between healthy plants and different diseases. The mildew severity stages in watermelon utilizing aerial
fact that there were so few misclassifications shows that the and ground remote sensing and machine
model was successful in picking up and extrapolating the key learning. Frontiers in Plant Science, 13, 791018.
characteristics from the training set. Visual similarities [3]. Abioye, E.A., Hensel, O., Esau, T.J., Elijah,
between disease symptoms, variances in image quality, or O., Abidin, M.S.Z., Ayobami, A.S. et al.
environmental factors such illumination discrepancies could (2022) Precision irrigation management using
be the cause of misclassifications that did occur. Even though machine learning and digital farming
the performance is great overall, expanding the dataset's size solutions. AgriEngineering, 4, 70–103.
and variety as well as investigating more sophisticated [4]. Akbar, M., Ullah, M., Shah, B., Khan, R.U., Hussain,
network topologies could be future enhancements. These T., Ali, F. et al. (2022) An effective deep learning
kinds of enhancements could greatly lower the approach for the classification of bacteriosis in peach
misclassification rates and enhance the model's generalisation leaves. Frontiers in Plant Science, 13, 1064854.
to unknown data. The model's ability to function well [5]. AlArfaj, A.A., Altamimi, A., Aljrees, T., Basheer,
suggests that it has potential for use in actual agricultural S., Umer, M., Samad, M.A. et al. (2023) Multi-step
settings where crop management and yield rely on early and preprocessing with UNet segmentation and transfer
accurate disease diagnosis. A dependable automated system learning model for pepper bell leaf disease
could promote food security by assisting farmers in taking detection. IEEE Access, 11, 132254–132267.
preventive action more effectively and minimising crop loss. [6]. Alberto, R.T., Rivera, J.C.E., Biagtan, A.R. & Isip,
Its utility could be increased even more by incorporating such M.F. (2020) Extraction of onion fields infected by
a model into automated agricultural machinery or mobile anthracnose-twister disease in selected municipalities
platforms. All things considered, the experiment effectively of Nueva Ecija using UAV imageries. Spatial
demonstrates the potential of deep learning in plant disease Information Research, 28, 383–389.
diagnosis, providing a significant advancement for precision [7]. Alirezazadeh, P., Schirrmann, M. & Stolzenburg,
farming and technological farming solutions. F. (2023) Improving deep learning-based plant disease
classification with attention mechanism. Gesunde
VI. CONCLUSION Pflanz, 75, 49–59.
[8]. Bao, D., Zhou, J., Bhuiyan, S.A., Zia, A., Ford,
This experiment effectively illustrated how machine R. & Gao, Y. (2021) Early detection of sugarcane
learning methods may use visual data to reliably identify and smut disease in hyperspectral images. In: In: 2021
categorise plant diseases. The confusion matrix heatmap 36th international conference on image and vision
shows that the model obtained excellent levels of accuracy computing New Zealand (IVCNZ). New York, USA:
with few misclassifications across many categories. These IEEE (Institute of Electrical and Electronics
findings imply that deep learning models can successfully Engineers), pp. 1–6.
identify intricate patterns in the symptoms of plant diseases, [9]. Bayer, P.E. & Edwards, D. (2021) Machine learning
offering a useful instrument for agricultural management and in agriculture: from silos to marketplaces. Plant
early diagnosis. The few mistakes found point to areas that Biotechnology Journal, 19, 648–650.
could use improvement, such adding additional data,
improving the quality of the images, or using more
sophisticated model architectures. However, the model's
good performance suggests that it might be used in the real
world, where prompt and precise disease identification is
essential to reducing crop losses and guaranteeing sustainable
food production. Future research can concentrate on
incorporating the system into precision farming platforms or
mobile applications to increase the effectiveness and
accessibility of disease monitoring. All things considered,
this study represents a positive step towards technologically
advanced, intelligent solutions in contemporary agriculture.

IJISRT25APR2394 www.ijisrt.com 3252