🌸 Flower Detection with Deep Learning (SE4050) 🌸

IT21302480 - Dilshan.O.A.P.

IT21195570 - Herath H.M.K.C.B

IT21298912 - Yapa Y.M.T.N.S

IT19985428 - Rathnasiri K.D.M.M

This project implements Flower Detection using deep learning models: ResNet152, VGG16, EfficientNet, and Xception.

🏗️ DL Project Overview

We trained models on a flower dataset consisting of 5 flower classes:

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

Models used:

1. ResNet152 (152 layers)
2. VGG16 (16 layers)
3. EfficientNet
4. Xception

---

🏋️‍♂️ Model Training

To train any model, ensure the dataset is in place, then use this general command structure:

# Example for training ResNet model
model = ResNet152(weights='imagenet', include_top=False, input_shape=(224, 224, 3))

# Common layers for all models
x = model.output
x = GlobalAveragePooling2D()(x)
x = Dense(128, activation='relu')(x)
predictions = Dense(5, activation='softmax')(x)

# Model creation
final_model = Model(inputs=model.input, outputs=predictions)

# Compile and train
final_model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])
final_model.fit(train_data, epochs=10, validation_data=val_data)

🧪 Evaluation

Evaluate the model:

accuracy = final_model.evaluate(test_data)[1] * 100
print(f"Test Accuracy: {accuracy}%")

🔮 Prediction

Make predictions on a new image:

img = image.load_img('/path/to/image.jpg', target_size=(224, 224))
img_array = np.expand_dims(image.img_to_array(img), axis=0)
prediction = final_model.predict(img_array)
print(f"Predicted Class: {np.argmax(prediction)}")

💾 Save & Load

Save the model:

final_model.save('flower_model.h5')

Load the model:

model = load_model('flower_model.h5')

🎯 Results

Expected accuracy for each model:

Model	Accuracy (%)
ResNet152	~96.5
VGG16	~94.2
EfficientNet	~97.1
Xception	~96.0

These results were obtained by training each model on the flower dataset for 10 epochs.

🌸 Happy Flower Detection!

This is the best compact version, focusing only on essential content for training, evaluating, and predicting flower detection.