A Deep CNN–DenseNet169 Architecture for Real-Time Plant Disease Detection and Classification

1.

CMR TECHNICAL CAMPUS
UGC(Autonomous)
Department of CSE (Data Science)
Project Stage -1
Review-1
A Deep CNN–DenseNet169 Architecture for Real-Time Plant
Disease Detection and Classification
Under Guidance:
Dr. P.V.NARESH
Associate Professor
A SAI SANKEERHT GOUD
A VIGNESH REDDY
K JASWANTH SAIKUMAR
N AISHVARYA REDDY
237R1A67D0
237R1A67D1
237R1A67F7
237R1A67H3

2.

VISION AND MISSION
VISION
To Provide quality education and research environment in Data Science that fosters
innovation and technological advancement to meet global challenges
MISSION
1.To provide an industry-aligned data science curriculum to develop strong analytical and
problem-solving skills.
2.To foster a research-driven environment that promotes innovation and technological
advancements.
3.To promote lifelong learning, teamwork and ethics for global challenges.

3.

Table of contents
❑ Abstract
❑ Problem Statement
❑ Objectives
❑ Literature Review
❑ Existing System
❑ Proposed System
❑ Methodology
❑ References

4.

Abstract
Plant diseases significantly reduce crop yield and quality, affecting agricultural
productivity worldwide. Early detection is essential to prevent disease spread and
improve crop management. This project proposes a deep learning-based approach that
combines a Custom Convolutional Neural Network (CNN) with DenseNet169 for
accurate plant disease detection and classification. The Custom CNN extracts low-level
image features, while DenseNet169 performs deep feature learning for improved
classification accuracy. The system processes leaf images in real time and classifies
their diseased categories. Experimental results demonstrate that the proposed model
achieves high accuracy with reduced computational complexity, making it suitable for
smart agriculture applications.
Keywords : Convolutional Neural Network(CNN),DenseNet169, Leaf Image Classification,
Plant Disease Detection, Real-Time Disease Detection.

5.

Problem Statement
➢ Agricultural Impact: Plant diseases significantly reduce crop yield, quality,
and overall plant health.
➢ Limitations of Traditional Methods: Manual disease identification is timeconsuming, costly, and may be affected by human errors.
➢ Need for Early Detection: Early symptoms of plant diseases are often difficult
to identify with the naked eye, making early diagnosis challenging.
➢ Proposed Solution: A Deep CNN–DenseNet169 model is used to
automatically analyze plant leaf images and classify plants as healthy
or diseased.
➢ Benefits: The system provides fast, accurate and automated disease
detection, helping farmers take timely preventive and corrective actions
to reduce crop losses.

6. Objectives

➢ To develop an automated system for detecting plant diseases using leaf images.
➢ To accurately classify plant diseases based on leaf images.
➢ To use a Custom CNN for extracting important low-level features such as
edges, shapes, and textures.
➢ To integrate DenseNet169 for deep feature extraction and improved disease
classification.
➢ To enable real-time disease detection from uploaded or captured leaf images.

7.

Literature Review
s.no
Title
1
AI Based Real-Time Disease Diagnosis in
Plants Using Deep Learning Driven CNNs
2
Medicinal Plant Leaf Disease
Classification Using Optimal Weighted
Features with Dilated Adaptive DenseNet
and Attention Mechanism
3
Enhanced Rice Disease Classification
through Pre-trained DenseNet Model
Using Transfer Learning
Author
Journal
Name &
Year
Key Findings
Gaps
D. Devarajan, R.
Allafi, M. Obayya
et al.
Scientific
Reports,
2026
Demonstrates the potential of CNNs for
accurate and real-time plant disease
detection.
Does not specifically combine a custom
CNN with DenseNet169 for feature
extraction and classification. (Nature)
R. Leelavathi, M.
Kalamani
Scientific
Reports,
2025
Demonstrated the effectiveness of
DenseNet-based deep feature extraction for
plant disease classification.
Complex architecture and optimization
increase computational requirements; realtime deployment is not the main focus.
(Nature)
Authors et al.
Franklin
Open, 2025
Achieved 96% accuracy in classifying four
rice-leaf disease categories.
Focused only on rice diseases; uses
DenseNet201 rather than DenseNet169
and does not provide a general multi-crop
solution. (ScienceDirect)
4
Comparing Pre-trained Models for
Efficient Leaf Disease Detection: A Study
on Custom CNN
Authors et al.
Journal of
Electrical
Systems
2024
Compared Custom CNN with several pretrained models, including DenseNet201,
EfficientNet, ResNet and VGG16 for leaf
disease detection.
Does not specifically combine Custom
CNN with DenseNet169 for sequential
low-level and deep feature extraction.
5
Multiclass Classification of Diseased
Grape Leaf Identification Using Deep
Convolutional Neural Network (DCNN)
Classifier
K. Vinayaka Prasad,
Hanumesh Vaidya,
Choudhari
Rajashekhar et al.
Scientific
Reports,
2024
Used CNN and DCNN based on VGG16 for
multiclass grape leaf disease classification
and achieved 99.06% test accuracy.
Focuses only on grape leaf diseases and
uses VGG16 rather than DenseNet169; it
does not combine Custom CNN and
DenseNet169.

8.

Existing Statement
➢ Two-phase frameworks use CNN+SNN for event detection followed by ResNet9for classification.
➢ Requires a large amount of labeled training data.
➢ Performance is affected by lighting and weather conditions in field imagery.
➢ ResNet-9 model is computationally heavy for edge deployment.
➢ Requires human supervision and regular maintenance.

9. Proposed Statement

➢ Develop a hybrid deep learning model using Custom CNN + DenseNet169.
➢ Preprocess leaf images by resizing, normalizing, and applying data augmentation.
➢ Use the Custom CNN to extract important leaf features such as color, texture, and
disease spots.
➢ Pass the extracted features to DenseNet169 for accurate disease classification.
➢ Use a Soft max layer to predict the plant disease category.
➢ Display the disease name and confidence score in real time.

10. System Architecture

11. Methodology

➢ Image Collection: Collect healthy and diseased plant leaf images from the
Plant Village dataset.
➢ Preprocessing: Resize images, normalize pixel values, and apply data
augmentation (rotation, flipping, zooming).
➢ Feature Extraction: Use a Custom CNN to extract important features such as
color, texture, and disease patterns.
➢ Classification: Pass the extracted features to DenseNet169 to classify the plant
disease.
➢ Prediction: Use a Soft max layer to predict the disease class with a confidence
score.
➢ Evaluation: Measure model performance using Accuracy, Precision, Recall,
and .
➢ Real-Time Detection: Deploy the trained model to detect plant diseases from
new leaf images in real time.

12.

References
➢ Devarajan, D., Allafi, R., Obayya, M., et al. (2026). “AI Based Real-Time Disease
Diagnosis in Plants Using Deep Learning Driven CNNs.” Scientific Reports.
➢ Leelavathi, R., & Kalamani, M. (2025). “Medicinal Plant Leaf Disease Classification
Using Optimal Weighted Features with Dilated Adaptive DenseNet and Attention
Mechanism.” Scientific Reports.
➢ “Enhanced Rice Disease Classification through Pre-trained DenseNet Model Using
Transfer Learning.” (2025). Franklin Open.
➢ Prasad, K. V., Vaidya, H., Rajashekhar, C., et al. (2024). “Multiclass Classification of
Diseased Grape Leaf Identification Using Deep Convolutional Neural Network
(DCNN) Classifier.” Scientific Reports.
➢ “Comparing Pre-trained Models for Efficient Leaf Disease Detection: A Study on
Custom CNN.” (2024). Journal of Electrical Systems and Information Technology.

13.

Thank you