Delete garbage_classification_preprocessing.ipynb

garbage_classification_preprocessing.ipynb

@@ -1,436 +0,0 @@
-{
- "cells": [
-  {
-   "cell_type": "markdown",
-   "id": "6e68dd4c",
-   "metadata": {},
-   "source": [
-    "# Garbage Classification — Full notebook with image preprocessing (step-by-step)\n",
-    "\n",
-    "This notebook:\n",
-    "- Loads the garbage classification dataset (from folder structure),\n",
-    "- Applies an **image preprocessing pipeline** (histogram equalization, blur, Sobel edges) using a `preprocessing_function` for `ImageDataGenerator`,\n",
-    "- Trains two models (DenseNet121 and ResNet101V2) with transfer learning,\n",
-    "- Evaluates and visualizes results (confusion matrix, classification report),\n",
-    "- Includes visualization of **original vs enhanced** images.\n",
-    "\n",
-    "**Note:** Adjust `path` if your dataset location is different (example: Google Drive or Kaggle dataset).\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "578daf0e",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 1) Imports\n",
-    "import numpy as np\n",
-    "import pandas as pd\n",
-    "import matplotlib.pyplot as plt\n",
-    "import seaborn as sns\n",
-    "import os\n",
-    "from PIL import Image\n",
-    "\n",
-    "import tensorflow as tf\n",
-    "from tensorflow.keras.layers import Conv2D , MaxPooling2D , Dense , Flatten , Dropout , GlobalAveragePooling2D\n",
-    "from tensorflow.keras.models import Sequential , Model\n",
-    "from tensorflow.keras.applications import DenseNet121, ResNet101V2\n",
-    "from pathlib import Path\n",
-    "from sklearn.model_selection import train_test_split\n",
-    "from sklearn.metrics import confusion_matrix, classification_report\n",
-    "\n",
-    "# optimizer\n",
-    "from tensorflow.keras.optimizers import Adam, AdamW\n",
-    "\n",
-    "import cv2\n",
-    "\n",
-    "import warnings\n",
-    "warnings.filterwarnings('ignore')\n",
-    "\n",
-    "print('TensorFlow version:', tf.__version__)\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "c507e2dd",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 2) Parameters & dataset path\n",
-    "# Change this path if your dataset is elsewhere (e.g., Google Drive or local)\n",
-    "path = '/kaggle/input/garbage-classification/garbage_classification'  # <-- keep or update\n",
-    "img_size = 128\n",
-    "batch_size = 32\n",
-    "random_state = 42\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "43490050",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 3) Build dataframe of image filepaths and labels\n",
-    "filepaths = []\n",
-    "labels = []\n",
-    "\n",
-    "for root, dirs, files in os.walk(path):\n",
-    "    for file in files:\n",
-    "        if file.lower().endswith(('.png', '.jpg', '.jpeg', '.gif', '.bmp')):\n",
-    "            filepath = os.path.join(root, file)\n",
-    "            filepaths.append(filepath)\n",
-    "            label = os.path.basename(root)\n",
-    "            labels.append(label)\n",
-    "\n",
-    "data_df = pd.DataFrame({'filepath': filepaths, 'original_label': labels})\n",
-    "print('Total images found:', len(data_df))\n",
-    "data_df.head()\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "7d26c28c",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 4) Unify labels (example: unify various glass subfolders into 'glass')\n",
-    "def unify_glass_labels(label):\n",
-    "    if 'glass' in label.lower():\n",
-    "        return 'glass'\n",
-    "    return label\n",
-    "\n",
-    "data_df['unified_label'] = data_df['original_label'].apply(unify_glass_labels)\n",
-    "data_df.drop(columns=['original_label'], inplace=True)\n",
-    "\n",
-    "print('Class distribution:')\n",
-    "display(data_df['unified_label'].value_counts())\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "c77a46c6",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 5) Train / Test split (stratified)\n",
-    "train_df, test_df = train_test_split(\n",
-    "    data_df,\n",
-    "    test_size=0.2,\n",
-    "    stratify=data_df['unified_label'],\n",
-    "    random_state=random_state\n",
-    ")\n",
-    "\n",
-    "print('Train samples:', len(train_df))\n",
-    "print('Test samples :', len(test_df))\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "370a47a7",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 6) Preprocessing function\n",
-    "# This function is compatible with ImageDataGenerator.preprocessing_function.\n",
-    "# Keras calls the preprocessing_function after rescale (so input here will be float32 in [0,1]).\n",
-    "# We convert back to 0-255 before applying cv2 operations, then return a float array in [0,1].\n",
-    "\n",
-    "def enhance_preprocessing(img):\n",
-    "    import numpy as np\n",
-    "    import cv2\n",
-    "    # img: float32 in [0,1], shape (H, W, 3), color order: RGB\n",
-    "    # Convert to uint8 [0,255]\n",
-    "    arr = (img * 255).astype('uint8')\n",
-    "    # Convert RGB -> Grayscale\n",
-    "    gray = cv2.cvtColor(arr, cv2.COLOR_RGB2GRAY)\n",
-    "    # Histogram equalization\n",
-    "    equalized = cv2.equalizeHist(gray)\n",
-    "    # Gaussian blur\n",
-    "    blurred = cv2.GaussianBlur(equalized, (3, 3), 0)\n",
-    "    # Sobel edge detection\n",
-    "    sobelx = cv2.Sobel(blurred, cv2.CV_64F, 1, 0, ksize=3)\n",
-    "    sobely = cv2.Sobel(blurred, cv2.CV_64F, 0, 1, ksize=3)\n",
-    "    sobel = np.sqrt(sobelx**2 + sobely**2)\n",
-    "    sobel = np.clip(sobel, 0, 255).astype('uint8')\n",
-    "    # Stack back to 3 channels (RGB-like)\n",
-    "    final = np.stack([sobel, sobel, sobel], axis=-1)\n",
-    "    # Convert to float [0,1]\n",
-    "    final = final.astype('float32') / 255.0\n",
-    "    return final\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "6e296335",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 7) Create ImageDataGenerators (with augmentation for training)\n",
-    "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
-    "\n",
-    "train_gen = ImageDataGenerator(\n",
-    "    rescale=1./255,\n",
-    "    rotation_range=20,\n",
-    "    width_shift_range=0.2,\n",
-    "    height_shift_range=0.2,\n",
-    "    shear_range=0.2,\n",
-    "    zoom_range=0.2,\n",
-    "    horizontal_flip=True,\n",
-    "    fill_mode='nearest',\n",
-    "    preprocessing_function=enhance_preprocessing  # apply our enhancement\n",
-    ")\n",
-    "\n",
-    "test_gen = ImageDataGenerator(\n",
-    "    rescale=1./255,\n",
-    "    preprocessing_function=enhance_preprocessing  # apply same preprocessing for evaluation\n",
-    ")\n",
-    "\n",
-    "train_data = train_gen.flow_from_dataframe(\n",
-    "    train_df,\n",
-    "    x_col='filepath',\n",
-    "    y_col='unified_label',\n",
-    "    target_size=(img_size, img_size),\n",
-    "    batch_size=batch_size,\n",
-    "    class_mode='categorical',\n",
-    "    color_mode='rgb',\n",
-    "    shuffle=True\n",
-    ")\n",
-    "\n",
-    "test_data = test_gen.flow_from_dataframe(\n",
-    "    test_df,\n",
-    "    x_col='filepath',\n",
-    "    y_col='unified_label',\n",
-    "    target_size=(img_size, img_size),\n",
-    "    batch_size=batch_size,\n",
-    "    class_mode='categorical',\n",
-    "    color_mode='rgb',\n",
-    "    shuffle=False\n",
-    ")\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "848d62b5",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 8) Class indices & labels\n",
-    "class_indices = train_data.class_indices\n",
-    "print('Class indices (label -> index):')\n",
-    "print(class_indices)\n",
-    "\n",
-    "# Build index -> label mapping for predictions later\n",
-    "index_to_label = {v: k for k, v in class_indices.items()}\n",
-    "classes = [index_to_label[i] for i in range(len(index_to_label))]\n",
-    "print('\\nClasses (in model index order):', classes)\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "2d37b6f8",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 9) Visualize: Original vs Enhanced\n",
-    "from tensorflow.keras.preprocessing.image import load_img, img_to_array\n",
-    "\n",
-    "# Pick a sample image from test_df\n",
-    "sample_fp = test_df['filepath'].iloc[0]\n",
-    "print('Sample filepath:', sample_fp)\n",
-    "\n",
-    "orig = img_to_array(load_img(sample_fp, target_size=(img_size, img_size))) / 255.0\n",
-    "enh = enhance_preprocessing(orig)\n",
-    "\n",
-    "fig, axes = plt.subplots(1,2, figsize=(10,5))\n",
-    "axes[0].imshow(orig)\n",
-    "axes[0].set_title('Original (rescaled)')\n",
-    "axes[0].axis('off')\n",
-    "\n",
-    "axes[1].imshow(enh)\n",
-    "axes[1].set_title('Enhanced (preprocessing_function)')\n",
-    "axes[1].axis('off')\n",
-    "plt.show()\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "693d4c04",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 10) Train DenseNet121 (transfer learning)\n",
-    "num_classes = len(class_indices)\n",
-    "\n",
-    "base_model = DenseNet121(input_shape=(img_size, img_size, 3), include_top=False, weights='imagenet')\n",
-    "base_model.trainable = True\n",
-    "\n",
-    "x = base_model.output\n",
-    "x = GlobalAveragePooling2D()(x)\n",
-    "x = Dropout(0.3)(x)\n",
-    "predictions = Dense(num_classes, activation='softmax')(x)\n",
-    "\n",
-    "model_DenseNet121 = Model(inputs=base_model.input, outputs=predictions)\n",
-    "\n",
-    "optimizer = AdamW(learning_rate=1e-4)\n",
-    "model_DenseNet121.compile(loss='categorical_crossentropy', optimizer=optimizer, metrics=['accuracy'])\n",
-    "model_DenseNet121.summary()\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "f7c970e1",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 11) Fit DenseNet121\n",
-    "from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint\n",
-    "\n",
-    "earlystop = EarlyStopping(patience=5, restore_best_weights=True, monitor='val_accuracy')\n",
-    "checkpoint_path = 'dense121_best.h5'\n",
-    "mc = ModelCheckpoint(checkpoint_path, monitor='val_accuracy', save_best_only=True, verbose=1)\n",
-    "\n",
-    "epochs = 20\n",
-    "\n",
-    "history_dense = model_DenseNet121.fit(\n",
-    "    train_data,\n",
-    "    validation_data=test_data,\n",
-    "    epochs=epochs,\n",
-    "    callbacks=[earlystop, mc]\n",
-    ")\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "46f10b3e",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 12) Evaluate DenseNet121\n",
-    "loss, accuracy = model_DenseNet121.evaluate(test_data)\n",
-    "print(f'DenseNet121 -> Loss: {loss:.4f}, Accuracy: {accuracy:.4f}')\n",
-    "\n",
-    "# Predictions\n",
-    "preds = model_DenseNet121.predict(test_data, verbose=1)\n",
-    "y_pred_idx = np.argmax(preds, axis=1)\n",
-    "y_pred_labels = [index_to_label[i] for i in y_pred_idx]\n",
-    "y_true_labels = test_df['unified_label'].values\n",
-    "\n",
-    "# Confusion matrix\n",
-    "plt.figure(figsize=(10,8))\n",
-    "cm = confusion_matrix(y_true_labels, y_pred_labels, labels=classes)\n",
-    "sns.heatmap(cm, annot=True, fmt='d', cmap='Blues', xticklabels=classes, yticklabels=classes)\n",
-    "plt.title('DenseNet121 - Confusion Matrix')\n",
-    "plt.show()\n",
-    "\n",
-    "print('\\nClassification Report:')\n",
-    "print(classification_report(y_true_labels, y_pred_labels, target_names=classes))\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "8968c73c",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 13) Train ResNet101V2 (transfer learning)\n",
-    "base_model = ResNet101V2(input_shape=(img_size, img_size, 3), include_top=False, weights='imagenet')\n",
-    "base_model.trainable = True\n",
-    "\n",
-    "x = base_model.output\n",
-    "x = GlobalAveragePooling2D()(x)\n",
-    "x = Dropout(0.5)(x)\n",
-    "predictions = Dense(num_classes, activation='softmax')(x)\n",
-    "\n",
-    "model_ResNet101V2 = Model(inputs=base_model.input, outputs=predictions)\n",
-    "\n",
-    "optimizer = AdamW(learning_rate=1e-4)\n",
-    "model_ResNet101V2.compile(loss='categorical_crossentropy', optimizer=optimizer, metrics=['accuracy'])\n",
-    "model_ResNet101V2.summary()\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "639722e9",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 14) Fit ResNet101V2\n",
-    "checkpoint_path_r = 'resnet101v2_best.h5'\n",
-    "mc_r = ModelCheckpoint(checkpoint_path_r, monitor='val_accuracy', save_best_only=True, verbose=1)\n",
-    "earlystop_r = EarlyStopping(patience=5, restore_best_weights=True, monitor='val_accuracy')\n",
-    "\n",
-    "history_resnet = model_ResNet101V2.fit(\n",
-    "    train_data,\n",
-    "    validation_data=test_data,\n",
-    "    epochs=epochs,\n",
-    "    callbacks=[earlystop_r, mc_r]\n",
-    ")\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "e2ba64ab",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 15) Evaluate ResNet101V2\n",
-    "loss_r, accuracy_r = model_ResNet101V2.evaluate(test_data)\n",
-    "print(f'ResNet101V2 -> Loss: {loss_r:.4f}, Accuracy: {accuracy_r:.4f}')\n",
-    "\n",
-    "# Predictions\n",
-    "preds_r = model_ResNet101V2.predict(test_data, verbose=1)\n",
-    "y_pred_idx_r = np.argmax(preds_r, axis=1)\n",
-    "y_pred_labels_r = [index_to_label[i] for i in y_pred_idx_r]\n",
-    "\n",
-    "plt.figure(figsize=(10,8))\n",
-    "cm_r = confusion_matrix(y_true_labels, y_pred_labels_r, labels=classes)\n",
-    "sns.heatmap(cm_r, annot=True, fmt='d', cmap='Blues', xticklabels=classes, yticklabels=classes)\n",
-    "plt.title('ResNet101V2 - Confusion Matrix')\n",
-    "plt.show()\n",
-    "\n",
-    "print('\\nClassification Report (ResNet101V2):')\n",
-    "print(classification_report(y_true_labels, y_pred_labels_r, target_names=classes))\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "b70f9f9e",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 16) Plot training history (DenseNet121 vs validation)\n",
-    "def plot_history(h, title='Model'):\n",
-    "    plt.figure(figsize=(8,4))\n",
-    "    plt.plot(h.history['accuracy'], label='train_acc')\n",
-    "    plt.plot(h.history['val_accuracy'], label='val_acc')\n",
-    "    plt.xlabel('Epoch')\n",
-    "    plt.ylabel('Accuracy')\n",
-    "    plt.legend()\n",
-    "    plt.title(title)\n",
-    "    plt.show()\n",
-    "\n",
-    "plot_history(history_dense, title='DenseNet121 Accuracy')\n",
-    "plot_history(history_resnet, title='ResNet101V2 Accuracy')\n"
-   ]
-  }
- ],
- "metadata": {},
- "nbformat": 4,
- "nbformat_minor": 5
-}