Paminder002/Explainable_AI

1#Data Preprocessing2 3import gradio as gr4from tensorflow.keras.models import Model5import matplotlib.pyplot as plt6import cv2  # For color mapping the heatmap7 8 9 10 11 12 13 14 15import os16import numpy as np17import tensorflow as tf18from tensorflow.keras.preprocessing.image import ImageDataGenerator19from PIL import Image20 21# Set image size and batch size22IMAGE_SIZE = (224, 224)23BATCH_SIZE = 3224 25# Paths to your dataset26TRAIN_PATH = 'archive'27 28# Data generator for loading and preprocessing images29datagen = ImageDataGenerator(rescale=1./255, validation_split=0.15)30 31train_data = datagen.flow_from_directory(32    TRAIN_PATH,33    target_size=IMAGE_SIZE,34    batch_size=BATCH_SIZE,35    class_mode='binary',36    subset='training'  # Set as training data37)38 39val_data = datagen.flow_from_directory(40    TRAIN_PATH,41    target_size=IMAGE_SIZE,42    batch_size=BATCH_SIZE,43    class_mode='binary',44    subset='validation'  # Set as validation data45)46 47#CNN Model Setup (Transfer Learning)48 49import tensorflow as tf50from tensorflow.keras.applications import ResNet5051from tensorflow.keras.layers import Dense, GlobalAveragePooling2D, Dropout52from tensorflow.keras.models import Model53 54# Define the input shape55input_shape = (224, 224, 3)56 57# Load ResNet50 with input shape and without the top layer58base_model = ResNet50(weights='imagenet', include_top=False, input_shape=input_shape)59 60# Freeze the layers in the base model61base_model.trainable = False62 63# Add custom layers on top64x = base_model.output65x = GlobalAveragePooling2D()(x)66x = Dense(128, activation='relu')(x)67x = Dropout(0.5)(x)68predictions = Dense(1, activation='sigmoid')(x)69 70# Define the model71model = Model(inputs=base_model.input, outputs=predictions)72 73# Compile the model74model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])75 76# Model summary77model.summary()78 79#Training the Model80 81# Train the model82history = model.fit(83    train_data,84    validation_data=val_data,85    epochs=10,  # Adjust epochs as needed86    verbose=187)88 89import matplotlib.pyplot as plt90 91# Plot the training and validation accuracy92plt.figure(figsize=(12, 6))93 94# Accuracy plot95plt.subplot(1, 2, 1)96plt.plot(history.history['accuracy'], label='Training Accuracy')97plt.plot(history.history['val_accuracy'], label='Validation Accuracy')98plt.title('Model Accuracy')99plt.xlabel('Epoch')100plt.ylabel('Accuracy')101plt.legend(loc='lower right')102plt.grid(True)103 104# Loss plot105plt.subplot(1, 2, 2)106plt.plot(history.history['loss'], label='Training Loss')107plt.plot(history.history['val_loss'], label='Validation Loss')108plt.title('Model Loss')109plt.xlabel('Epoch')110plt.ylabel('Loss')111plt.legend(loc='upper right')112plt.grid(True)113 114# Show the plot115plt.tight_layout()116plt.show()117 118#Explainable AI Integration (Grad-CAM)119 120import numpy as np121import tensorflow as tf122import matplotlib.pyplot as plt123from tensorflow.keras.models import Model124from PIL import Image125 126def make_gradcam_heatmap(img_array, model, last_conv_layer_name):127    grad_model = Model(128        inputs=[model.inputs],129        outputs=[model.get_layer(last_conv_layer_name).output, model.output]130    )131 132    # Record operations for automatic differentiation133    with tf.GradientTape() as tape:134        conv_outputs, predictions = grad_model(img_array)135        loss = predictions[:, 0]  # Assuming binary classification (0 = Healthy, 1 = COVID-19)136 137    # Compute gradients138    grads = tape.gradient(loss, conv_outputs)139    pooled_grads = tf.reduce_mean(grads, axis=(0, 1, 2))140 141    conv_outputs = conv_outputs[0]142    heatmap = tf.reduce_mean(tf.multiply(pooled_grads, conv_outputs), axis=-1)143    heatmap = np.maximum(heatmap, 0) / np.max(heatmap)  # Normalize between 0 and 1144    return heatmap145 146def display_gradcam(img_path, heatmap, alpha=0.4):147    img = Image.open(img_path)148    img = img.resize((224, 224))  # Resize the image to match model input size149 150    heatmap = np.uint8(255 * heatmap)  # Convert heatmap to 0-255 scale151    heatmap = Image.fromarray(heatmap).resize((img.size), Image.LANCZOS)152    heatmap = np.array(heatmap)153 154    # Create figure to plot the image and heatmap155    fig, ax = plt.subplots(1, 2, figsize=(10, 5))156    ax[0].imshow(img)157    ax[1].imshow(img)158    ax[1].imshow(heatmap, cmap='jet', alpha=alpha)  # Overlay the heatmap159    plt.show()160 161# Load and preprocess the image162def preprocess_image(image_path):163    img = Image.open(image_path)164    img = img.resize((224, 224))  # Resize to match the input shape of the model165    img = np.array(img) / 255.0   # Normalize pixel values between 0 and 1166    img = np.expand_dims(img, axis=0)  # Add batch dimension167    return img168 169# Path to the image170img_path = 'archive/dataset/covid/01E392EE-69F9-4E33-BFCE-E5C968654078.jpeg'171 172# Preprocess the image173img_array = preprocess_image(img_path)174 175# Get the heatmap176heatmap = make_gradcam_heatmap(img_array, model, 'conv5_block3_out')  # Replace with your last conv layer's name177 178# Display the original image with the Grad-CAM heatmap overlay179display_gradcam(img_path, heatmap)180 181#Evaluation182 183# Evaluate model on validation data184test_loss, test_acc = model.evaluate(val_data, verbose=2)185print(f'Test Accuracy: {test_acc:.2f}')186 187 188# UI for the model189 190 191import gradio as gr192import numpy as np193from PIL import Image194import tensorflow as tf195from tensorflow.keras.models import Model196import matplotlib.pyplot as plt197import cv2  # For color mapping the heatmap198 199# Define the Grad-CAM function200def make_gradcam_heatmap(img_array, model, last_conv_layer_name):201    grad_model = Model([model.inputs], [model.get_layer(last_conv_layer_name).output, model.output])202    with tf.GradientTape() as tape:203        conv_outputs, predictions = grad_model(img_array)204        loss = predictions[:, 0]  # For binary classification205    grads = tape.gradient(loss, conv_outputs)206    pooled_grads = tf.reduce_mean(grads, axis=(0, 1, 2))207    conv_outputs = conv_outputs[0]208    heatmap = tf.reduce_mean(tf.multiply(pooled_grads, conv_outputs), axis=-1)209    heatmap = np.maximum(heatmap, 0)  # ReLU activation to make it non-negative210    heatmap = heatmap / np.max(heatmap)  # Normalize between 0 and 1211    return heatmap212 213# Function to overlay the heatmap on the original image214def apply_heatmap_to_image(img, heatmap):215    # Resize heatmap to match image size216    heatmap = cv2.resize(heatmap, (img.size[0], img.size[1]))217 218    # Convert heatmap to RGB (apply 'jet' colormap)219    heatmap_colored = cv2.applyColorMap(np.uint8(255 * heatmap), cv2.COLORMAP_JET)220 221    # Convert to RGB mode (since OpenCV uses BGR)222    heatmap_colored = cv2.cvtColor(heatmap_colored, cv2.COLOR_BGR2RGB)223 224    # Overlay the heatmap on the original image225    overlay = np.array(img) * 0.6 + heatmap_colored * 0.4226    overlay = np.clip(overlay, 0, 255).astype('uint8')227    return Image.fromarray(overlay)228 229# Define the prediction and explainability function230def predict_and_explain(img):231    img = Image.fromarray(img).resize((224, 224))  # Resize image for the model232    img_array = np.array(img) / 255.0   # Normalize pixel values233    img_array = np.expand_dims(img_array, axis=0)  # Add batch dimension234 235    # Get the prediction236    prediction = model.predict(img_array)237    confidence = float(prediction[0][0])238    result = "COVID-19 Positive" if confidence > 0.5 else "Healthy"239 240    # Generate the Grad-CAM heatmap241    last_conv_layer_name = 'conv5_block3_out'  # Update with the actual last convolution layer name242    heatmap = make_gradcam_heatmap(img_array, model, last_conv_layer_name)243 244    # Apply heatmap on the image245    heatmap_img = apply_heatmap_to_image(img, heatmap)246 247    # Display confidence and heatmap248    confidence_text = f"Confidence: {confidence:.2f}"249    return result, confidence_text, heatmap_img250 251# Gradio interface252def create_interface():253    gr_interface = gr.Interface(254        fn=predict_and_explain,255        inputs=gr.Image(type="numpy"),256        outputs=[gr.Textbox(label="Prediction"), gr.Textbox(label="Confidence"), gr.Image(label="Heatmap")],257        title="COVID-19 X-ray Classification with Explainability",258        description="Upload an X-ray image to predict if the patient has COVID-19, see the confidence score, and view the Grad-CAM heatmap."259    )260    return gr_interface261 262# Launch the interface263gr_interface = create_interface()264gr_interface.launch()