Quantum-Based Plant Disease Detector

A hybrid Classical–Quantum Machine Learning project for detecting and classifying tomato plant diseases from leaf images. The project combines Convolutional Neural Network (CNN) feature extraction, Principal Component Analysis (PCA), and a Variational Quantum Circuit (VQC) to investigate the potential of quantum machine learning for plant disease classification.

---

Project Overview

Plant diseases can significantly affect crop productivity and quality. Early and accurate identification of diseases can help farmers take timely preventive measures.

This project develops an intelligent image-based plant disease detection system that:

• Processes tomato leaf images.
• Extracts meaningful visual features using MobileNetV2.
• Reduces the high-dimensional feature space using PCA.
• Encodes the reduced features into a 4-qubit quantum circuit.
• Uses a Variational Quantum Circuit (VQC) for classification.
• Implements a classical SVM classifier as a baseline.
• Compares the classical and quantum approaches.

The primary objective is to explore whether a quantum-enhanced classifier can provide a useful alternative to conventional machine learning approaches for plant disease classification.

---

Objectives

1. Detect diseases in tomato plant leaves using image-based classification.
2. Extract deep visual features using a pretrained CNN.
3. Reduce feature dimensionality using Principal Component Analysis.
4. Encode the reduced features into quantum states.
5. Develop a Variational Quantum Circuit for classification.
6. Develop a classical SVM model for comparison.
7. Evaluate and compare the performance of the classical and quantum approaches.

---

Methodology

The complete workflow follows a hybrid classical–quantum pipeline:

Tomato Leaf Images
        ↓
Dataset Preparation
        ↓
CNN Feature Extraction
(MobileNetV2)
        ↓
Feature Scaling
        ↓
PCA Dimensionality Reduction
        ↓
4 Principal Components
        ↓
Angle Embedding
        ↓
4-Qubit Variational Quantum Circuit
        ↓
Disease Classification

A parallel classical pipeline is used for comparison:

CNN Features
     ↓
Feature Scaling
     ↓
SVM Classifier
     ↓
Disease Prediction

---

Technologies Used

Technology	Purpose
Python	Core programming language
TensorFlow / Keras	CNN and MobileNetV2
NumPy	Numerical computation and data storage
Scikit-learn	PCA, SVM, preprocessing and evaluation
PennyLane	Quantum circuit simulation and VQC
MobileNetV2	Deep feature extraction
SVM	Classical baseline classifier

---

Dataset

The project uses tomato leaf images representing multiple healthy and diseased conditions.

The dataset contains the following classes:

1. bacterial_spot
2. early_blight
3. healthy
4. late_blight
5. leaf_mold
6. septoria_leaf_spot
7. spider_mites
8. target_spot
9. yellow_leaf_curl_virus
10. mosaic_virus

The dataset is organized into training and testing splits.

---

CNN Feature Extraction

A pretrained MobileNetV2 model with ImageNet weights is used as the feature extractor.

The final classification layer is removed and global average pooling is used to obtain a compact feature representation of each tomato leaf image.

model = tf.keras.applications.MobileNetV2(
    weights="imagenet",
    include_top=False,
    pooling="avg"
)

model.trainable = False

This allows the project to use powerful pretrained visual representations without training a complete CNN from scratch.

---

PCA Dimensionality Reduction

CNN features are high-dimensional and cannot be directly mapped onto a small quantum circuit efficiently.

Therefore, the extracted features are:

1. Standardized using StandardScaler.
2. Reduced to 4 principal components using PCA.

pca = PCA(n_components=4)

X_train_pca = pca.fit_transform(X_train_scaled)
X_test_pca = pca.transform(X_test_scaled)

The four PCA components are then used as the four input features for the quantum circuit.

---

Variational Quantum Circuit

The quantum component uses a 4-qubit Variational Quantum Circuit implemented using PennyLane.

Quantum Pipeline

4 PCA Features
      ↓
Angle Embedding
      ↓
RY Trainable Rotations
      ↓
CNOT Entanglement
      ↓
Pauli-Z Measurements
      ↓
Classification Output

Quantum Circuit Components

1. Angle Embedding

The four PCA features are encoded into four qubits using Y-axis rotations.

qml.AngleEmbedding(
    features,
    wires=range(n_qubits),
    rotation="Y"
)

2. Trainable RY Rotations

Each qubit contains a trainable rotation:

qml.RY(weights[i], wires=i)

3. Entanglement

Neighbouring qubits are connected using CNOT gates:

qml.CNOT(wires=[i, i + 1])

4. Measurement

The expectation value of Pauli-Z is measured from each qubit.

These measurements form the output of the quantum model.

---

Classical Baseline

To evaluate the usefulness of the quantum approach, a classical Support Vector Machine (SVM) classifier with an RBF kernel is implemented.

model = SVC(
    kernel="rbf",
    probability=True,
    random_state=42
)

The classical model uses the CNN features after standardization and predicts the disease class for each test image.

The following metrics are calculated:

• Accuracy
• Precision
• Recall
• F1-Score

This provides a baseline against which the quantum model can be studied.

---

Model Comparison

The project is designed to compare:

Model	Feature Representation	Classifier
Classical Model	CNN Features	SVM
Quantum Model	4 PCA Components	Variational Quantum Circuit

The comparison helps investigate the feasibility of using a compact quantum circuit for image classification after classical feature extraction and dimensionality reduction.

---

Project Structure

Project-Exhibition/
│
├── data/
│   └── Original dataset
│
├── splits/
│   ├── train/
│   │   ├── bacterial_spot/
│   │   ├── early_blight/
│   │   ├── healthy/
│   │   ├── late_blight/
│   │   ├── leaf_mold/
│   │   ├── septoria_leaf_spot/
│   │   ├── spider_mites/
│   │   ├── target_spot/
│   │   └── yellow_leaf_curl_virus/
│   │
│   └── test/
│
├── cnn_features.py
├── cnn_test.py
├── check_dataset.py
│
├── train_classical.py
├── classical_model.py
│
├── train_quantum.py
├── vqc_test.py
│
├── X_train_pca.npy
├── X_test_pca.npy
├── y_train.npy
└── y_test.npy

---

Installation

Clone the repository:

git clone https://github.com/Jhanvisrivastava2107/Project-Exhibition.git
cd Project-Exhibition

Install the required Python libraries:

pip install numpy tensorflow scikit-learn pennylane

---

Running the Project

1. Extract CNN Features

Run the CNN feature extraction pipeline:

python cnn_features.py

The extracted features are stored for subsequent processing.

2. Prepare PCA Data

Run:

python train_classical.py

This performs:

• Train-test splitting
• Feature scaling
• PCA dimensionality reduction
• Saving of PCA datasets

3. Train the Classical Model

Run:

python classical_model.py

This trains the SVM classifier and reports:

• Accuracy
• Precision
• Recall
• F1-score

4. Train the Quantum Model

Run:

python train_quantum.py

This:

• Loads the PCA features.
• Builds the 4-qubit quantum circuit.
• Optimizes the trainable quantum parameters.
• Generates predictions on the test set.
• Saves the quantum predictions and trained weights.

5. Test the Quantum Circuit

The circuit structure can also be inspected using:

python vqc_test.py

---

Key Features

• Tomato plant disease classification
• Deep feature extraction using MobileNetV2
• PCA-based dimensionality reduction
• 4-qubit Variational Quantum Circuit
• Quantum feature entanglement using CNOT gates
• Classical SVM baseline
• Performance evaluation using multiple metrics
• Classical vs Quantum machine learning comparison

---

Future Scope

The project can be further improved by:

• Training and evaluating deeper or fine-tuned CNN architectures.
• Experimenting with different quantum feature-encoding techniques.
• Increasing the number of quantum layers and trainable parameters.
• Testing different quantum optimizers.
• Running the circuit on real quantum hardware.
• Performing extensive hyperparameter tuning.
• Comparing VQC with additional classical classifiers.
• Developing a user-friendly web or mobile interface for real-time disease detection.
• Providing treatment or prevention recommendations based on the detected disease.

---

Team Members

Name	Registration Number
Tejal Singh	25BCE10060
Chahak Gupta	25BCE10138
Jahnavi Gautam	25BCE10361
Jhanvi Srivastava	25BCE10405
Monisha Khare	25BCE11318

---

Project Information

Project Type: Project Exhibition / Academic Project Domain: Artificial Intelligence & Quantum Machine Learning Application: Plant Disease Detection Approach: Hybrid Classical–Quantum Machine Learning

---

Conclusion

This project demonstrates a hybrid approach to plant disease detection by combining deep learning-based feature extraction with dimensionality reduction and quantum machine learning.

By using MobileNetV2 for extracting meaningful image features and PCA for reducing them to four dimensions, the resulting representations can be efficiently processed by a 4-qubit Variational Quantum Circuit.

A classical SVM model is also implemented to provide a baseline for evaluating the effectiveness of the quantum approach. The project therefore serves as an exploration of how quantum machine learning can be integrated with conventional computer vision pipelines for practical classification problems.

---

Acknowledgement

This project was developed as part of an academic Project Exhibition, with the objective of exploring the application of emerging technologies such as Quantum Machine Learning in real-world agricultural problems.