© 2024 JETIR December 2024, Volume 11, Issue 12 www.jetir.

org (ISSN-2349-5162)

KrishiCare @ AI-based Plant Disease Detection

and Fertilizer Recommendation System using
Image Classification
Innovative Method for AI-based Plant Disease Detection and Fertilizer Recommendation
System using Image Classification
1
Dr. M.K. Jayanthi Kannan, 2Om Kumar Singh , 3Roshmik Agrawal ,
4
Aadish Chaturvedi, 5Harshit Tiwari , 6Abhyanand Sharma
1
Professor, 2,3,4,5,6 Student, School of Computing Science Engineering and Artificial Intelligence, VIT Bhopal University, Bhopal-
Indore Highway, Kothrikalan, Sehore, Madhya Pradesh - 466114

Abstract: Agriculture is a critical pillar of global food security but faces persistent threats from plant diseases that cause significant
crop losses and economic instability. Traditional methods of diagnosing and managing plant diseases are often slow, inaccurate,
and reliant on expert intervention, posing challenges for smallholder farmers. This paper introduces Krishi Care, an AI-powered
system for plant disease detection and management, designed to provide accurate and timely solutions. The system uses deep
learning models to identify plant diseases by analyzing leaf images and offers actionable recommendations tailored to the detected
issue. The system incorporates a recommendation engine to guide farmers on the appropriate use of fertilizers, pesticides, and
preventive measures, with real-time weather integration ensuring context-aware advice. The solution is accessible through an
intuitive web interface, empowering farmers and agricultural professionals to make informed decisions with minimal technical
expertise. Trained on the PlantVillage dataset, the system demonstrates an accuracy exceeding 96%, highlighting its potential for
scalable, efficient, and sustainable agricultural management. Future developments aim to expand its functionality by incorporating
multilingual support, IoT integration, and dataset generalization, making it adaptable to diverse farming contexts.

Index Terms - Plant Disease Detection, AI in Agriculture, Recommendation System, Sustainable Farming, Context-Aware
Solutions, Streamlit.

I. INTRODUCTION
Agriculture is a cornerstone of global food security and economic development, yet it faces persistent challenges from plant diseases that
significantly reduce crop yields and increase production costs. These diseases pose a severe threat to farmers, especially smallholders in rural
and underserved areas, where traditional methods of disease identification are often inaccessible, time-consuming, and reliant on expert
intervention. Delayed diagnosis and incorrect treatment further exacerbate the problem, resulting in substantial economic losses and
diminished agricultural productivity. Advancements in artificial intelligence (AI) and machine learning (ML) have revolutionized the potential
for addressing these challenges. AI-powered systems that integrate image classification with actionable insights provide scalable and efficient
solutions for early disease detection and management. With the growing adoption of smartphones and digital tools, such technologies are
becoming increasingly accessible, even in remote regions. This paper introduces Krishi Care, an integrated system that combines deep learning
for plant disease detection with a recommendation engine offering tailored advice on fertilizers, pesticides, and preventive measures. The
inclusion of real-time weather guidance ensures context-aware recommendations, making the system practical and effective. By blending
cutting-edge technology with a user-friendly interface, Krishi Care seeks to empower farmers, reduce crop losses, and promote sustainable
farming practices.

JETIR2412111 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org b191
© 2024 JETIR December 2024, Volume 11, Issue 12 www.jetir.org (ISSN-2349-5162)

II. LITERATURE REVIEW

Table 1: Literature Review of KrishiCare

S.No. Study Objective Results Techniques Used Datasets

Plant Disease Detection, Detect crop Over 92%
1 CNN PlantVillage
2016 diseases accuracy

Revolutionizing agriculture
with artificial intelligence: Revolutionizing
Accuracies over CNN (ResNet, VGG,
2 plant disease detection plant disease PlantVillage
90% and MobileNet)
methods, applications, and detection with AI
their limitations, 2024

High accuracy CNNs, Faster R- PlantVillage,

Plant diseases and pests’ Automate plant (95-99%) across CNN, YOLO, U-Net, PlantDoc,
3 detection based on deep disease detection various datasets data augmentation, Rice Leaf
learning, 2021 using CNNs and transfer learning, and Diseases,
methodologies segmentation PDDB

50.8% accuracy Natural Language

A field-based recommender 100
(recommendation Processing (NLP),
system for crop disease Real-time disease smallholder
4 system), 88% Convolutional Neural
detection using machine detection farmers in
accuracy (disease Networks (CNN),
learning, 2023 Uganda
detection) MobileNetV3

Mean Average Kaggle Plant

Detect apple
A Deep Learning Enabled Precision: 91.2%, Pathology
diseases with YOLOv4, DenseNet,
5 Multi-Class Plant Disease F1 score: 95.9%, Apple Dataset
better accuracy SPP, Mish activation.
Detection Model, 2021 +9.05% in augmented
and speed.
precision. tenfold.

• Plant Disease Detection

Develop an effective system for the automated detection of plant diseases using deep learning models, specifically convolutional neural
networks (CNNs). The primary goal was to create a scalable model that could classify plant diseases from images with high accuracy.
The authors highlight the importance of early and accurate disease detection in agriculture to improve crop yield and reduce losses.
Traditional disease diagnosis methods are often time-consuming, labor-intensive, and require expert knowledge, making it challenging
to implement at scale. (Mohanty et al. 2016). The study utilized a large dataset consisting of 54,306 images covering 38 different plant
diseases across 14 crop species. The images were split into training and testing sets to build and evaluate the deep learning model. The
authors employed a pre-trained deep learning architecture (AlexNet and GoogleNet) to classify the images and fine-tuned it for the
plant disease classification task. Their deep learning model achieved an impressive accuracy of over 92%, demonstrating that CNNs
can be effectively used for plant disease detection. The study also discussed the potential real-world applications of such a system,
including the development of mobile applications that could help farmers identify plant diseases in the field using just a smartphone.
The authors conclude that deep learning presents a promising tool for improving global food security by enhancing disease management
in crops.
• Revolutionizing agriculture with artificial intelligence: plant disease detection methods, applications, and their
limitations
The article explores how artificial intelligence could revolutionize contemporary agriculture, with a particular emphasis on plant
disease detection. AI-driven approaches are replacing conventional disease identification methods, which frequently call for a
significant investment of time and expertise. To produce quicker and more precise results, these automated techniques make use of
processes like image acquisition, preprocessing, segmentation, feature extraction, and classification. It has been established that
ML and DL models, in particular, CNNs are very useful for classifying and diagnosing diseases in a variety of crops, including
cucumbers, tomatoes, potatoes, and chilies. ResNet, VGG, and MobileNet are examples of pre-trained architectures that have
demonstrated exceptional performance, frequently attaining classification accuracies of over 90%. The study highlights how crucial
it is to use curated databases and open datasets like PlantVillage in order to create reliable training models. These datasets include
annotated photos of both healthy and diseased leaves that have been enhanced using methods such as data augmentation. Common
disease-specific symptoms, like mosaic patterns, blight, and leaf spots, are well documented. The AI frameworks discussed in the
paper have increased the dependability of disease detection systems by concentrating on data preprocessing to lower noise and
improve image quality. Emerging technologies like smart drones and Internet of Things (IoT) devices are also covered in the paper
as ways to improve disease detection and real-time monitoring at the field level. In spite of these developments, the study outlines
a number of restrictions that prevent the widespread application of AI in agriculture. They point out that it is made worse by dataset
JETIR2412111 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org b192
© 2024 JETIR December 2024, Volume 11, Issue 12 www.jetir.org (ISSN-2349-5162)

inconsistencies, ambient noise, and the need for a lot of processing power when processing images at high resolution in real time
and face challenges due to different lighting conditions and capture angles. These difficulties highlight the need for additional
algorithmic developments to strengthen and bolster AI systems in practical settings. The study offers a number of potential avenues
for further investigation. Creating scalable systems with real-time monitoring capabilities is one of the main goals of integrating AI
with IoT. Developing more extensive datasets that cover a range of ailments and uncommon diseases is another top priority since
it would increase the generalizability of the models. Lastly, computational efficiency is critical to ensure that these systems remain
accessible and affordable, especially for small-scale farmers. (Abbas Jafar, et al, 2024)

• Plant diseases and pests’ detection based on deep learning

The research paper Liu and Wang (2021) investigates the use of deep learning for plant disease and pest detection, considering the
crucial challenges that surround agriculture. Traditional methods mainly rely on manual feature designs and are based on strong
knowledge. They tend to fail when applied in highly complex or natural settings. The study highlights how deep learning,
particularly with CNNs, has revolutionized the detection of plant diseases by automating the extraction of features, thus improving
both accuracy and reliability. The review classifies the methods into three types: classification networks, such as AlexNet and
ResNet; detection networks, which include Faster R-CNN and YOLO; and segmentation networks, including U-Net and Mask R-
CNN. Such techniques exhibit high accuracy in detecting, localizing, and segmenting diseased areas, even under challenging
conditions. Although widely used datasets such as PlantVillage abound, the authors claim that there is a large limitation because of
the under availability of diverse datasets, which can accurately represent a natural environment. Further, the paper identifies key
challenges including small sample sizes, the need for detailed recognition, early disease detection, and performance issues tied to
occlusion and fluctuating lighting conditions. Suggested solutions in the paper include data augmentation, transfer learning, and
better network architectures. Particular emphasis is placed upon the need for lightweight models for real-time deployment in
resource-constrained devices as applied in agricultural settings. For the future, efforts should be made so comprehensive datasets
develop while using multispectral data to enhance the early detection of diseases while integrating insights from a multidisciplinary
approach in framing effective agricultural solutions. According to the study, the potential of deep learning does present considerable
scope to be able to advance crop management and ensure global food security but calls for constant innovation and collaboration.
(Liu and Wang et al. 2021)
• A field-based recommender system for crop disease detection using ML
The article reports on a practical solution to the problem for smartphone-based applications powered by artificial intelligence in
support of detection and management of crop diseases. Losses to farmers from pests and diseases are enormous, with an annual
yield of 35–40% lost through pests and diseases in rural areas. Such expertise is often unavailable or delayed, hence their reliance
on outdated or inequitable practice. This app offers real-time diagnostic feedback and farming recommendations through a complex
combination of ML and NLP. In short, this will put the power in farmers to act quickly and improve their harvest (Jonathan Omara,
et al, 2023). It analyzes images of crops taken by the camera phone, diagnoses common diseases on crops like cassava mosaic and
brown streak, and gives advice for managing such a situation. It uses a model called RetBERT, which has been specifically
developed with the objective to answer questions raised by farmers, compared to a database of expert-provided information to
generate their responses. The app can function both online and offline, even where internet connectivity is lacking, especially in
remote areas. To contribute to the building of this knowledge base, farmers shared their experiences and questions. The dataset
hence acquired is nearly 4,000 question-answer pairs. Tests proved the effectiveness of models with AI implemented in the
application. The most accurate and efficient support is offered by the RetBERT model, and the function of great disease detection
possessed a 88% accuracy level. Today, farmers can self-diagnose conditions that mean no waiting period and less dependence
upon a visit by agricultural experts. The product may actually enhance food security in the region because it can lead to timely
intervention and improved farming practices. The researchers suggest further improvement of the system using additional audio
and video inputs and the database of crops for more crops, making the system a strong versatile agriculture assistant for farmers
anywhere on earth.

• A Deep Learning Enabled Multi-Class Plant Disease Detection Model Based on Computer Vision
In this paper, Arunabha M. Roy and Jayabrata Bhaduri propose a deep learning model YOLOv4 that is used for real-time detection
of plant diseases in apples. The model increases the architecture of YOLOv4 by dense blocks for better feature sharing,
CSPDarkNet53 for improved feature extraction, and Mish activation function to increase the accuracy of the training. The model
includes Spatial Pyramid Pooling (SPP) for an enlarged receptive field and Path Aggregation Network (PANet) to effectively fuse
multi-scale features. It attained excellent performance, reaching a mAP of 91.2% and F1-score of 95.9%, thereby increasing the
precision of detection by 9.05% compared with the standard YOLOv4 model. The researchers applied the Kaggle Plant Pathology
Apple Dataset, which they extended to mimic diverse real-world conditions. This model presents huge possibilities for the
accurate detection of scab and rust-type diseases in apple crops, making it an extremely useful tool in precision agriculture.
(Arunabha M. Roy and Jayabrata Bhaduri, 2021)
III. KRISHICARE @ AI-BASED PLANT DISEASE DETECTION PROPOSED SYSTEM

The proposed model for plant disease detection is built around a custom Convolutional Neural Network (CNN) architecture, tailored
for efficient image classification into various disease categories. Instead of relying on pre-trained architectures like MobileNetV2
or ResNet50, the model is lightweight and specifically designed for this task. It processes input images of size 256×256×3 through
convolutional layers with ReLU activation for feature extraction, MaxPooling layers for dimensionality reduction, a Flatten layer
to convert 2D feature maps into a 1D vector, and fully connected Dense layers for classification into categories such as "Healthy,"
"Powdery," and "Rust." The model concludes with a SoftMax activation function for multi-class classification. The model is
compiled using the Adam optimizer and categorical cross-entropy loss, ensuring compatibility with multi-class classification tasks.

JETIR2412111 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org b193
© 2024 JETIR December 2024, Volume 11, Issue 12 www.jetir.org (ISSN-2349-5162)

Training involves pre-processed datasets and validation to monitor performance, optimizing for high accuracy and low loss. A user-
friendly interface, developed using Streamlit, enhances accessibility, allowing users to upload plant images and receive disease
predictions seamlessly. This system is straightforward and effective, balancing simplicity with functionality. Its modularity and
scalability make it adaptable for similar image classification tasks. The integration of Streamlit ensures usability for non-technical
users, particularly in agricultural settings where real-time plant disease detection can have a significant impact.

IV. ARCHITECTURE DIAGRAM OF KRISHICARE @ AI-BASED PLANT DISEASE DETECTION

Figure 1: Architecture Diagram

V. PROJECT FUNCTIONAL MODULES IMPLEMENTATION

The Plant Disease Detection System efficiently utilizes essential libraries for numerical computations, visualization, and deep
learning, including NumPy for numerical operations, plotly for interactive visualizations, and keras and TensorFlow for defining
and training CNN models with layers like Conv2D, MaxPooling2D, Flatten, and Dense, optimized for GPU usage. The dataset is
organized into directories for training, validation, and testing, with subdirectories for specific disease labels such as "Healthy,"
"Powdery," and "Rust." Preprocessing ensures uniformity through image resizing to (350x250) pixels using Pillow, normalization
of pixel values to a [0, 1] range, and visualization tools for dataset inspection. Hardware optimization is achieved by leveraging
GPUs with TensorFlow memory growth and distributed strategies to enhance training performance. Data analysis scripts verify
dataset integrity, count images in categories, and address inconsistencies. The CNN model architecture comprises convolutional
layers for feature extraction, pooling layers for dimensionality reduction, and fully connected layers for classification. Model
training involves using training data for learning, validation data for tuning hyperparameters, and testing data for performance
evaluation, with metrics like accuracy and loss visualized through graphs. Finally, the trained model is deployed to classify plant
images and predict diseases, providing a scalable and adaptable solution for plant health monitoring and similar classification
problems.

JETIR2412111 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org b194
© 2024 JETIR December 2024, Volume 11, Issue 12 www.jetir.org (ISSN-2349-5162)

VI. KrishiCare PROTOTYPE, ALGORITHM AND PROGRAM LOGIC

Figure 2: Building the Convolutional Neural Network Logic of KrishiCare

Figure 3: KrishiCare Training and Testing the Model

JETIR2412111 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org b195
© 2024 JETIR December 2024, Volume 11, Issue 12 www.jetir.org (ISSN-2349-5162)

Figure 4: KrishiCare Loss Over Epochs Graph

Figure 5: KrishiCare Confusion Matrix

JETIR2412111 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org b196
© 2024 JETIR December 2024, Volume 11, Issue 12 www.jetir.org (ISSN-2349-5162)

Figure 6: ROC Curves of KrishiCare

Figure 7: Model Loss and Model Accuracy Graph of KrishiCare

JETIR2412111 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org b197
© 2024 JETIR December 2024, Volume 11, Issue 12 www.jetir.org (ISSN-2349-5162)

Figure 8: Sample of a leaf image from dataset and saving the mode of KrishiCare

Figure 9: KrishiCare Loss and Accuracy Over Epochs Graph

JETIR2412111 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org b198
© 2024 JETIR December 2024, Volume 11, Issue 12 www.jetir.org (ISSN-2349-5162)

Figure 10: Working of the App KrishiCare

VII. METHODOLOGY FOR DEVELOPING THE KrishiCare SYSTEM

Krishi Care is developed through a structured methodology, starting with defining objectives and understanding user needs, such
as real-time feedback and ease of use. A diverse dataset of plant images, categorized by disease types, is collected and preprocessed
through resizing, normalization, and augmentation for consistency and robustness. A custom Convolutional Neural Network (CNN)
is designed with layers for feature extraction, dimensionality reduction, and classification, optimized using the Adam optimizer and
categorical cross-entropy loss. The model is trained and validated on a split dataset, with performance metrics monitored to refine
its accuracy. It is deployed using a Streamlit application, providing a user-friendly interface for uploading plant images and
receiving disease predictions with treatment recommendations. The system is rigorously tested with unseen data to ensure reliability
and hosted on a cloud platform for scalability and accessibility. Regular updates to the model and dataset enhance accuracy and
allow adaptation to new disease categories, delivering an efficient and scalable solution.

VIII. KrishiCare SYSTEMS FINDINGS AND RESULTS DISCUSSION

The plant disease detection project demonstrated a high level of accuracy in classifying plant diseases like "Healthy," "Powdery,"
and "Rust," using a custom Convolutional Neural Network (CNN). The system processed images efficiently, enabling real-time
predictions, and proved scalable for larger datasets. The integration of Streamlit provided a user-friendly interface, making the tool
accessible even to non-technical users such as farmers and agricultural researchers. The model performed robustly on unseen data,
ensuring reliability, while the disease-specific treatment recommendations added practical value for agricultural applications. The
system showed potential for real-world impact by reducing crop loss and improving productivity, and its modular design allows for
easy updates and future enhancements, such as incorporating new diseases or expanding to other plant species.

IX. CONCLUSION
The proposed system combines state-of-the-art AI techniques with user-centric design to address the challenges of plant disease
detection and management. By combining image-based diagnostics, expert recommendations, and weather insights, the system
empowers farmers to make informed decisions. Future work will involve expanding the disease dataset, incorporating soil analysis,
IoT integration, and multilingual support for wider adoption. The farmers can effectively diagnose disease conditions and less
dependence upon a visit by agricultural experts. The research project focuses on AI-based plant disease detection efficiently and
recommends fertilizers to overcome the issues and improve the health conditions based on image classification, to enhance food
security in the region because it can lead to timely intervention and improved farming practices.
REFERENCES
[1] Mohanty, S. P., Hughes, D. P., & Salathé, M. "Using Deep Learning for Image-Based Plant Disease Detection," Frontiers in
Plant Science, 2016.
[2] Balajee RM, Jayanthi Kannan MK, Murali Mohan V. Image-Based Authentication Security Improvement by Randomized
Selection Approach. InInventive Computation and Information Technologies 2022 (pp. 61-71). Springer, Singapore.
[3] Eric Ries, The Lean Startup: How Today's Entrepreneurs Use Continuous Innovation to Create Radically Successful Businesses, 1st
edition, Crown Business, 2011.
[4] Kavitha, E., Tamilarasan, R., Baladhandapani, A., Kannan, M.K.J. (2022). A novel soft clustering approach for gene expression
data. Computer Systems Science and Engineering, 43(3), 871-886. https://doi.org/10.32604/csse.2022.021215
[5] Robert C. Martin, Clean Code: A Handbook of Agile Software Craftsmanship, 1st edition, Prentice Hall, 2008.
JETIR2412111 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org b199
© 2024 JETIR December 2024, Volume 11, Issue 12 www.jetir.org (ISSN-2349-5162)

[6] G., D. K., Singh, M. K., & Jayanthi, M. (Eds.). (2016). Network Security Attacks and Countermeasures. IGI Global.
https://doi.org/10.4018/978-1-4666-8761-5
[7] Agricultural Data Insights, "Common Plant Diseases and Their Remedies," 2021.
[8] M. K. Jayanthi, "Strategic Planning for Information Security -DID Mechanism to befriend the Cyber Criminals to assure Cyber
Freedom," 2017 2nd International Conference on Anti-Cyber Crimes (ICACC), Abha, Saudi Arabia, 2017, pp. 142-147, doi:
10.1109/Anti-Cybercrime.2017.7905280.
[9] Sam Newman, Building Microservices: Designing Fine-Grained Systems, 1st edition, O'Reilly Media, 2015.
[10] Naik, Harish and Kannan, M K Jayanthi, A Survey on Protecting Confidential Data over Distributed Storage in Cloud
(December 1, 2020). Available at SSRN: https://ssrn.com/abstract=3740465 or http://dx.doi.org/10.2139/ssrn.3740465
[11] R M B, M K JK. Intrusion Detection on AWS Cloud through Hybrid Deep Learning Algorithm. Electronics. 2023; 12(6):1423.
https://doi.org/10.3390/electronics12061423
[12] Kavitha, E., Tamilarasan, R., Poonguzhali, N., Kannan, M.K.J. (2022). Clustering gene expression data through modified
agglomerative M-CURE hierarchical algorithm. Computer Systems Science and Engineering, 41(3), 1027-141.
https://doi.org/10.32604/csse.2022.020634
[13] Kumar, K.L.S., Kannan, M.K.J. (2024). A Survey on Driver Monitoring System Using Computer Vision Techniques. In:
Hassanien, A.E., Anand, S., Jaiswal, A., Kumar, P. (eds) Innovative Computing and Communications. ICICC 2024. Lecture
Notes in Networks and Systems, vol 1021. Springer, Singapore. https://doi.org/10.1007/978-981-97-3591-4_21

JETIR2412111 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org b200