🌸 Flower Image Classification with CNN

Project Description

This project trains a CNN model to recognize and classify 5 different types of flowers:

• 🌼 Daisy
• 🌾 Dandelion
• 🌹 Rose
• 🌻 Sunflower
• 🌷 Tulip

Features

• ✅ Reading and preprocessing images
• ✅ Automatic resize of images to standard dimensions
• ✅ Normalize pixel values
• ✅ Data Augmentation to improve model generalization
• ✅ Training of Convolutional Neural Network (CNN)
• ✅ Automatic saving of the best model
• ✅ Early Stopping and Learning Rate Scheduling
• ✅ Drawing accuracy and loss graphs
• ✅ Predicting new images with graphical result display

Project Structure

Image-Classification/
│
├── dataset/
│   ├── train/              # Training data (5 subfolders for each class)
│   │   ├── daisy/
│   │   ├── dandelion/
│   │   ├── rose/
│   │   ├── sunflower/
│   │   └── tulip/
│   └── validation/         # Validation data
│       ├── daisy/
│       ├── ...
│
├── model.py                # CNN model architecture
├── train.py                # Training script
├── predict.py              # Prediction script
├── prepare_dataset.py      # Dataset preparation and splitting
├── requirements.txt        # Dependencies
├── README.md              
└── flower_model.h5         # Saved model (after training)

Technologies Used

• Python
• TensorFlow / Keras - Deep Learning Framework
• OpenCV - Image Processing
• NumPy - Numerical Computations
• Matplotlib - Plot Visualization

How to Use

1️⃣ Install Dependencies

pip install -r requirements.txt

2️⃣ Download and Prepare Dataset

Download the dataset from Kaggle: 🔗 Flowers Recognition Dataset

curl -L -o ~/flowers-recognition.zip\
  https://www.kaggle.com/api/v1/datasets/download/alxmamaev/flowers-recognition

After extraction, use the preparation script:

python prepare_dataset.py --source path/to/flowers --split 0.8

This command splits the dataset into train and validation with 80/20 ratio.

3️⃣ Train the Model

python train.py

Train Screenshots

Outputs:

• flower_model.h5 - Trained model
• training_history.png - Accuracy and loss graph
• class_indices.txt - Class mapping

4️⃣ Predict New Image

python predict.py path/to/image.jpg

Or with options:

python predict.py rose_sample.jpg --model flower_model.h5

Model Architecture

The designed CNN model includes 4 convolutional blocks:

Layer	Filters	Kernel	Activation
Conv2D + BN + MaxPool	32	3×3	ReLU
Conv2D + BN + MaxPool	64	3×3	ReLU
Conv2D + BN + MaxPool	128	3×3	ReLU
Conv2D + BN + MaxPool	128	3×3	ReLU
Flatten + Dropout(0.5)	-	-	-
Dense + BN + Dropout(0.3)	512	-	ReLU
Dense (Output)	5	-	Softmax

Optimizer: Adam (lr=0.001)
Loss: Categorical Crossentropy
Metrics: Accuracy

Results

With default settings and 30 epochs of training, the model usually reaches 80-88% accuracy on validation data.

⭐ If this project was useful to you, please give it a star!