LandCoverPrediction_DL.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "f1d49544-d0a8-45dc-9b2e-48f5d3f299f0",
   "metadata": {
    "id": "MFF_uXoaWwTT"
   },
   "source": [
    "# Project  : Landcover prediction using Deep learning\n",
    "## Shahab Esfandiar"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "72f9baf1-97be-4527-88ed-9b4cfaf6db6a",
   "metadata": {
    "id": "j_O92uiZWwTV"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f2e06c37-b790-4eb3-8d12-5dffb9045907",
   "metadata": {
    "id": "BHny79eCWwTW"
   },
   "source": [
    "#### *At first, we should import the required libraries such as Numpy, Tensorflow, Gdal, Matplotlib and so on*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "1e31d8a6-91ad-4d78-adef-9fd89274fa30",
   "metadata": {},
   "outputs": [],
   "source": [
    "import warnings\n",
    "warnings.filterwarnings('ignore')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "id": "0e969230-51c6-4d4a-9616-fdcdf2a4921b",
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import random\n",
    "import shutil\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import seaborn as sb\n",
    "from PIL import Image\n",
    "from osgeo import gdal\n",
    "import tensorflow as tf\n",
    "from tensorflow import keras\n",
    "import matplotlib.pyplot as plt\n",
    "from matplotlib.lines import Line2D\n",
    "from sklearn.metrics import f1_score\n",
    "from scipy.interpolate import interpn\n",
    "from skimage.util import view_as_windows\n",
    "from tensorflow.keras.layers import Dense \n",
    "from tensorflow.keras.models import Model\n",
    "from tensorflow.keras import regularizers\n",
    "from tensorflow.keras import backend as K \n",
    "from tensorflow.keras.layers import Conv2D  \n",
    "from tensorflow.keras.models import load_model\n",
    "from tensorflow.keras.layers import MaxPooling2D \n",
    "from tensorflow.keras.layers import TimeDistributed\n",
    "from tensorflow.keras.layers import Flatten, Reshape\n",
    "from tensorflow.keras.metrics import Precision, Recall\n",
    "from tensorflow.keras.layers import concatenate, Input\n",
    "from tensorflow.keras.layers import BatchNormalization\n",
    "from tensorflow.keras.layers import SimpleRNN, LSTM, GRU, Dropout\n",
    "from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7bb2cb68-7f5a-450d-879b-6d91cab634c1",
   "metadata": {
    "id": "oN0GABo0WwTZ"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4098e2b9-3d6b-445a-8924-206634106495",
   "metadata": {},
   "source": [
    "## <span style=\"color:green;\">1. Data Prepration</span>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "14055bdb",
   "metadata": {
    "id": "BHny79eCWwTW"
   },
   "source": [
    "#### *To prepare the dataset for this project, we needed pre-processing which was done using <span style=\"color:red;\">ArcGIS Pro</span>. These pre-processings are respectively :*"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d66d2afa-bd0b-4b96-809f-a385f7c3546c",
   "metadata": {},
   "source": [
    ">*1. Converting the coordinate system of population density rasters from WGS to Canadian Spatial Reference System.*\n",
    "\n",
    ">*2. Clipping the above rasters based on the case study area and land cover rasters.*\n",
    "\n",
    ">*3. Equalizing the pixel size of the rasters with each other and resampling them to 30 meter pixels.*"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7ed35be2-66e6-4fbc-a313-24294ab0bf85",
   "metadata": {},
   "source": [
    "#### *We did all the above steps in the export data section of the software and by applying the mentioned settings. Since the coordinate system of the population density rasters is based on the land cover raster for export. We fixed the error that occurred when clipping in arcgis software.*"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d7671ca2-db8e-48f1-9b2a-919d083597cd",
   "metadata": {},
   "source": [
    "#### *For this reason, the pro version of this software brings us fewer steps and higher accuracy than the previous version.*"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a70754c8-8256-4b6b-953a-9b80a8c4f484",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7280d9a7-8f49-403d-ad9e-4d98eaaf9435",
   "metadata": {},
   "source": [
    "#### *In order to prepare the dataset, in the next step, after equalizing their spatial resolution, we must also equalize their temporal resolution.*"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "94e83af3-0303-460f-8923-d1b6910c980d",
   "metadata": {},
   "source": [
    "#### *For this, we use a linear interpolation using meshgrid. so as to create an approximation using 5 existing rasters for the years 2000 to 2020 and save the approximate rasters. With this, all the spatial requirements for the dataset of this project are provided.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b523aa95-44c3-4fab-b829-1397a3dfe8ec",
   "metadata": {},
   "outputs": [],
   "source": [
    "Image.MAX_IMAGE_PIXELS = None\n",
    "output = \"D:\\GeoAi\\Week8\\Project\\PopDen\"\n",
    "\n",
    "image_2000 = Image.open(\"D:\\GeoAi\\Week8\\Project\\PopulationDensity\\p2000.tif\")\n",
    "image_2005 = Image.open(\"D:\\GeoAi\\Week8\\Project\\PopulationDensity\\p2005.tif\")\n",
    "image_2010 = Image.open(\"D:\\GeoAi\\Week8\\Project\\PopulationDensity\\p2010.tif\")\n",
    "image_2015 = Image.open(\"D:\\GeoAi\\Week8\\Project\\PopulationDensity\\p2015.tif\")\n",
    "image_2020 = Image.open(\"D:\\GeoAi\\Week8\\Project\\PopulationDensity\\p2020.tif\")\n",
    "\n",
    "arr = np.r_['0,3', image_2000, image_2005, image_2010, image_2015, image_2020]\n",
    "slices, rows, cols = arr.shape\n",
    "\n",
    "#Create meshgrid\n",
    "[X, Y, Z] = np.meshgrid(np.arange(cols), np.arange(rows), np.arange(slices))\n",
    "\n",
    "if not os.path.exists(output):\n",
    "    os.makedirs(output)\n",
    "\n",
    "for year in range(2001, 2020):\n",
    "    \n",
    "    #Determination of weights\n",
    "    [X2, Y2, Z2] = np.meshgrid(np.arange(cols), np.arange(rows), [(year - 2000) / 20 * 4])\n",
    "    Vi = interpn((X, Y, Z), arr, np.array([X2, Y2, Z2]).T)\n",
    "\n",
    "    plt.imshow(Vi[0], cmap=plt.cm.gray)\n",
    "    plt.savefig(os.path.join(output, f\"p{year}.tif\"))\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6fde38af",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "07653ec6-33f1-49a4-abd5-31fcc744fc70",
   "metadata": {
    "id": "aciiJUNgE5Sh"
   },
   "source": [
    "#### *Now we need to define our data format as specified for this project. Therefore, with the following code snippet, we will create the required folders for data storage for the years 2000 to 2020.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "5f96d3a5-49cd-4d50-ad17-da4a95e30573",
   "metadata": {},
   "outputs": [],
   "source": [
    "path = \"D:\\Learning\\GeoAi\\Week8\\Data\\Train\"\n",
    "\n",
    "for i in range(2001, 2020):\n",
    "    folder_name = \"y\" + str(i)\n",
    "    new_path = os.path.join(path, folder_name)\n",
    "    os.makedirs(new_path)\n",
    "\n",
    "    try:\n",
    "        os.makedirs(new_path)\n",
    "\n",
    "    except FileExistsError:\n",
    "        pass\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "58351caf",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8ca53003",
   "metadata": {
    "id": "aciiJUNgE5Sh"
   },
   "source": [
    "#### *At this stage, we need to extract the patch from the raster images of each year. Since the dimensions of the images are large, we convert each raster into a number of smaller 64 x 64 non-overlapping rasters.*"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9d69d65c",
   "metadata": {
    "id": "aciiJUNgE5Sh"
   },
   "source": [
    "#### *Then we have to remove the negative or nan values from our data and finally save the patches as two-dimensional arrays in the folder corresponding to each year. Also, since we have to use the data of 2019 as a test of our prediction model, we store its arrays in a separate path.*"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c014390b-2bca-4471-8b4e-ebac98a42672",
   "metadata": {
    "id": "aciiJUNgE5Sh"
   },
   "source": [
    "#### *After completing the above steps, pre-processing is needed. First of all, we have to encode the land cover data and convert it into 11 bands so that the model does not make mistakes in learning. Then we have to save all the arrays in Float32 format so that the required storage space is less.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9f0b8c9c-077c-4537-af29-3d99c2a6a275",
   "metadata": {},
   "outputs": [],
   "source": [
    "input_pop_dir = \"D:\\\\Learning\\\\GeoAi\\\\Week8\\\\Data\\\\PopDen\"\n",
    "input_lc_dir = \"D:\\\\Learning\\\\GeoAi\\\\Week8\\\\Data\\\\LandCover\"\n",
    "output_dir = \"D:\\\\Learning\\\\GeoAi\\\\Week8\\\\Data\\\\Train\"\n",
    "\n",
    "# Patch size and step for sliding window\n",
    "patch_size = (64, 64) \n",
    "patch_step = 64\n",
    "c = 2001\n",
    "\n",
    "# Loop through each year from 2001 to 2019\n",
    "for year in range(2001, 2020):\n",
    "\n",
    "    # Construct file paths for population density and land cover data\n",
    "    input_pop_file = os.path.join(input_pop_dir, f\"p{year}.tif\") \n",
    "    input_lc_file = os.path.join(input_lc_dir, f\"LC{year}.tif\") \n",
    "    output_subdir = os.path.join(output_dir, f\"y{year}\") \n",
    "    \n",
    "    # Create output directory if it doesn't exist\n",
    "    if not os.path.exists(output_subdir):\n",
    "        os.makedirs(output_subdir)\n",
    "    \n",
    "    # Open raster files\n",
    "    rasterPop_image = gdal.Open(input_pop_file)\n",
    "    rasterLC_image = gdal.Open(input_lc_file)\n",
    "    \n",
    "    # Read raster data as arrays\n",
    "    rasterPop_array = rasterPop_image.ReadAsArray()\n",
    "    rasterLC_array = rasterLC_image.ReadAsArray()\n",
    "\n",
    "    # Handle no-data values in population density array\n",
    "    rasterPop_array[rasterPop_array < 0] = np.nan\n",
    "    col_mean_Pop = np.nanmean(rasterPop_array, axis=0)\n",
    "    index_Pop = np.where(np.isnan(rasterPop_array))\n",
    "    rasterPop_array[index_Pop] = np.take(col_mean_Pop, index_Pop[1])\n",
    "    \n",
    "    # Extract patches using sliding window\n",
    "    pop_patches = view_as_windows(rasterPop_array, patch_size, step=patch_step) \n",
    "    lc_patches = view_as_windows(rasterLC_array, patch_size, step=patch_step) \n",
    "    \n",
    "    num_x = pop_patches.shape[0]\n",
    "    num_y = pop_patches.shape[1] \n",
    "    val = list(np.unique(rasterLC_array))\n",
    "    \n",
    "    # Loop through each patch\n",
    "    for i in range(num_x):\n",
    "        for j in range(num_y):\n",
    "            \n",
    "            # Construct output file path\n",
    "            output_file = os.path.join(output_subdir, f\"arr{year}_{i}_{j}.npz\") \n",
    "            pop_patch = pop_patches[i, j] \n",
    "            lc_patch = lc_patches[i, j] \n",
    "\n",
    "            # One-hot encode land cover data\n",
    "            x, y = lc_patch.shape\n",
    "            lc = np.zeros((x, y, 11), dtype=np.float32)\n",
    "\n",
    "            for k in range(11):\n",
    "                if np.sum(lc_patch == val[k]) > 0:\n",
    "                    X, Y = np.where(lc_patch == val[k])\n",
    "                    lc[X, Y, k] = 1\n",
    "\n",
    "            # Save patches as compressed .npz files\n",
    "            pop_patch = pop_patch.astype(np.float32)\n",
    "            lc = lc.astype(np.float32)\n",
    "            arr = np.concatenate((lc, pop_patch[..., np.newaxis]), axis=2)\n",
    "            arr = arr.astype(np.float32)\n",
    "            np.savez_compressed(output_file, arr=arr)\n",
    "\n",
    "    print(f\"Processed y{c}\")\n",
    "    c += 1\n",
    "            "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2eac9ad2-4bbb-46cd-b75c-96456f1dfb76",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "63c8a252-0cea-4087-881f-cb014ecac9e8",
   "metadata": {},
   "source": [
    "## <span style=\"color:green;\">2. Data Preprocessing</span>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5ca1faee-f80d-41b7-aea8-c6d80ba20f7a",
   "metadata": {
    "id": "aciiJUNgE5Sh"
   },
   "source": [
    "#### *With the completion of data preparation and pre-processing steps, it is time to create data in time series format according to the model and <span style=\"color:red;\">CNN-GRU</span> selected for this project.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f349966e-45df-47b1-bf31-95d1899d9e74",
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# Parameters for the data dimensions and time steps\n",
    "num_patches = 2500\n",
    "time_step, width, height, channel = 4, 64, 64, 12\n",
    "num_years = 19\n",
    "u = 0\n",
    "co = 1\n",
    "\n",
    "# Loop through each patch grid (50x50)\n",
    "for i in range(50):\n",
    "    for j in range(50):\n",
    "        c = 1\n",
    "        \n",
    "        # Loop through each year minus the time step\n",
    "        for k in range(num_years - time_step):\n",
    "            # Load the first array for the current year\n",
    "            first_array = np.load(os.path.join(\"D:\\\\Learning\\\\GeoAi\\\\Week8\\\\Data\\\\Train\", f\"y{2001+k}\", f\"arr{2001+k}_{i}_{j}.npz\"))['arr']\n",
    "            assert first_array.shape == (width, height, channel)\n",
    "            next_arrays = []\n",
    "            \n",
    "            # Load subsequent arrays for the time steps\n",
    "            for z in range(1, time_step + 1):\n",
    "                a = np.load(os.path.join(\"D:\\\\Learning\\\\GeoAi\\\\Week8\\\\Data\\\\Train\", f\"y{2001+k+z}\", f\"arr{2001+k+z}_{i}_{j}.npz\"))['arr']\n",
    "                next_arrays.append(a)\n",
    "            \n",
    "            # Ensure all arrays have the correct shape\n",
    "            for array in next_arrays:\n",
    "                assert array.shape == (width, height, channel)\n",
    "        \n",
    "            # Stack the arrays to create the input data\n",
    "            input = np.stack([first_array] + next_arrays[:-1], axis=0)\n",
    "            input = input.astype(np.float32)\n",
    "            np.savez_compressed(os.path.join(\"D:\\\\Learning\\\\GeoAi\\\\Week8\\\\Project\\\\Train\\\\Input\", f\"input_{u}.npz\"), input=input)\n",
    "            \n",
    "            # Prepare the label data by removing the last channel\n",
    "            label_train = np.delete(next_arrays[-1], -1, axis=2)\n",
    "            label_train = label_train.astype(np.float32)\n",
    "            np.savez_compressed(os.path.join(\"D:\\\\Learning\\\\GeoAi\\\\Week8\\\\Project\\\\Train\\\\Label\", f\"label_{u}.npz\"), label=label_train)\n",
    "            \n",
    "            # Print progress\n",
    "            print(f\"Processed sample {c} of patch {co} out of {num_patches}\")\n",
    "            c += 1\n",
    "            u += 1\n",
    "            \n",
    "        co += 1\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "37279b64-0633-4ac2-bc45-fa9148659c3b",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ea0b7516-aa58-4006-9237-b761aa754a33",
   "metadata": {},
   "source": [
    "#### *Now we need to separate the last 2500 arrays related to the year 2019 from the data and save them in another folder with a suitable name.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "id": "807ce164-ac03-457c-b018-872fa36c224a",
   "metadata": {},
   "outputs": [],
   "source": [
    "test_path = \"D:\\Learning\\GeoAi\\Week8\\Project\\Test\"\n",
    "\n",
    "for folder in [\"input\", \"label\"]:\n",
    "    folder_path = os.path.join(test_path, folder)\n",
    "\n",
    "    for file in os.listdir(folder_path):\n",
    "        \n",
    "        file_number = int(file.split(\"_\")[1].split(\".\")[0])\n",
    "        new_name = f\"{folder}_{file_number - 35000}.npz\"\n",
    "        os.rename(os.path.join(folder_path, file), os.path.join(folder_path, new_name))\n",
    "        "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2706d64b-9412-4606-b9af-64cf349ff125",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4ce9d8c7-0a2f-4a51-bf8d-3a007ac411d3",
   "metadata": {},
   "source": [
    "#### *Next, we should define the directory path for training/test data and list files in that directory.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "ec7c23cb-8043-40e5-ae0a-c69bb6351868",
   "metadata": {},
   "outputs": [],
   "source": [
    "input_loc_train = \"D:\\Learning\\GeoAi\\Week8\\Project\\Train\\Input\"\n",
    "input_files_train = os.listdir(input_loc_train)\n",
    "\n",
    "label_loc_train = \"D:\\Learning\\GeoAi\\Week8\\Project\\Train\\Label\"\n",
    "label_files_train = os.listdir(label_loc_train)\n",
    "\n",
    "input_loc_test = \"D:\\Learning\\GeoAi\\Week8\\Project\\Test\\Input\"\n",
    "input_files_test = os.listdir(input_loc_test)\n",
    "\n",
    "label_loc_test = \"D:\\Learning\\GeoAi\\Week8\\Project\\Test\\Label\"\n",
    "label_files_test = os.listdir(label_loc_test)\n",
    "\n",
    "# Create a dictionary to map input file names to corresponding label file names for training/test data\n",
    "image_label_map = {\"input_{}.npz\".format(i): \"label_{}.npz\".format(i)\n",
    "                    for i in range(len(input_files_train))}\n",
    "\n",
    "image_label_map_val = {\"input_{}.npz\".format(i): \"label_{}.npz\".format(i)\n",
    "                       for i in range(len(input_files_test))}\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2eef7423-be67-404a-9af0-dfcc358e010c",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b9890aec-d3d6-427a-a591-da1a8200e080",
   "metadata": {},
   "source": [
    "#### *Now we need to create two <span style=\"color:red;\">Data Generator</span> classes to avoid memory errors. Because we have a very large number of samples and the gpu memory is not capable of processing this amount of data. As a result, by using this class, we give the data to the model in the form of 4 so that our graphic memory has the capacity to process it.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "c2c61c09-c34c-4224-86c4-2cc571fd2b53",
   "metadata": {},
   "outputs": [],
   "source": [
    "time_step, width, height, channel = 4, 64, 64, 12\n",
    "\n",
    "# Define a custom data generator class for training/test data\n",
    "class DataGenerator(tf.keras.utils.Sequence):\n",
    "    def __init__(self, list_examples, batch_size = 4, dim = (time_step, width, height, channel), shuffle = True):\n",
    "        # Constructor of the data generator.\n",
    "        self.dim = dim\n",
    "        self.batch_size = batch_size\n",
    "        self.list_examples = list_examples\n",
    "        self.shuffle = shuffle\n",
    "        self.on_epoch_end()\n",
    "\n",
    "    def __len__(self):\n",
    "        # Denotes the number of batches per epoch\n",
    "        return int(np.floor(len(self.list_examples) / self.batch_size))\n",
    "\n",
    "    def __getitem__(self, index):\n",
    "        # Generate one batch of data\n",
    "        indexes = self.indexes[index * self.batch_size:(index + 1) * self.batch_size]\n",
    "\n",
    "        # Find list of IDs\n",
    "        list_IDs_temp = [self.list_examples[k] for k in indexes]\n",
    "\n",
    "        # Generate data\n",
    "        X, y = self.__data_generation(list_IDs_temp)\n",
    "\n",
    "        return X, y\n",
    "\n",
    "    def on_epoch_end(self):\n",
    "        # This function is called at the end of each epoch.\n",
    "        self.indexes = np.arange(len(self.list_examples))\n",
    "        if self.shuffle == True:\n",
    "            np.random.shuffle(self.indexes)\n",
    "\n",
    "    def __data_generation(self, list_IDs_temp):\n",
    "        # Load individual numpy arrays and aggregate them into a batch.\n",
    "\n",
    "        X = np.empty([self.batch_size, self.dim[0], self.dim[1], self.dim[2], self.dim[3]], dtype = np.float32)\n",
    "\n",
    "        # y is a one-hot encoded vector.\n",
    "        y = np.empty([self.batch_size, width, height, 11], dtype = np.float32)\n",
    "\n",
    "        # Generate data.\n",
    "\n",
    "        c = 0\n",
    "        for i in list_IDs_temp:\n",
    "\n",
    "            x_file_path = os.path.join(input_loc_train, i)\n",
    "            y_file_path = os.path.join(label_loc_train, image_label_map.get(i))\n",
    "\n",
    "            # Load sample\n",
    "            X[c, :, :, :, :] = np.load(x_file_path)['input']\n",
    "            \n",
    "            # Load labels    \n",
    "            y[c, :, :, :] = np.load(y_file_path)['label']\n",
    "            \n",
    "\n",
    "            c += 1\n",
    "\n",
    "        return X, y\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "dc9ba14b-538a-469d-a3a9-cdd742e14c45",
   "metadata": {},
   "outputs": [],
   "source": [
    "class ValDataGenerator(tf.keras.utils.Sequence):\n",
    "    def __init__(self, list_examples, batch_size = 4, dim = (time_step, width, height, channel), shuffle = True):\n",
    "        # Constructor of the data generator.\n",
    "        self.dim = dim\n",
    "        self.batch_size = batch_size\n",
    "        self.list_examples = list_examples\n",
    "        self.shuffle = shuffle\n",
    "        self.on_epoch_end()\n",
    "\n",
    "    def __len__(self):\n",
    "        # Denotes the number of batches per epoch\n",
    "        return int(np.floor(len(self.list_examples) / self.batch_size))\n",
    "\n",
    "    def __getitem__(self, index):\n",
    "        # Generate one batch of data\n",
    "        indexes = self.indexes[index * self.batch_size:(index + 1) * self.batch_size]\n",
    "\n",
    "        # Find list of IDs\n",
    "        list_IDs_temp = [self.list_examples[k] for k in indexes]\n",
    "\n",
    "        # Generate data\n",
    "        X, y = self.__data_generation(list_IDs_temp)\n",
    "\n",
    "        return X, y\n",
    "\n",
    "    def on_epoch_end(self):\n",
    "        # This function is called at the end of each epoch.\n",
    "        self.indexes = np.arange(len(self.list_examples))\n",
    "        if self.shuffle == True:\n",
    "            np.random.shuffle(self.indexes)\n",
    "\n",
    "    def __data_generation(self, list_IDs_temp):\n",
    "        # Load individual numpy arrays and aggregate them into a batch.\n",
    "\n",
    "        X = np.empty([self.batch_size, self.dim[0], self.dim[1], self.dim[2], self.dim[3]], dtype = np.float32)\n",
    "\n",
    "        # y is a one-hot encoded vector.\n",
    "        y = np.empty([self.batch_size, width, height, 11], dtype = np.float32)\n",
    "\n",
    "        # Generate data.\n",
    "\n",
    "        c = 0\n",
    "        for i in list_IDs_temp:\n",
    "\n",
    "            x_file_path = os.path.join(input_loc_test, i)\n",
    "            y_file_path = os.path.join(label_loc_test, image_label_map_val.get(i))\n",
    "\n",
    "            # Load sample\n",
    "            X[c, :, :, :, :] = np.load(x_file_path)['input_train']\n",
    "            # Load labels     \n",
    "            y[c, :, :, :] = np.load(y_file_path)['label_train']\n",
    "\n",
    "            c += 1\n",
    "\n",
    "        return X, y\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "id": "2d0c37c0-4a20-4feb-bf0f-e63496950e1f",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create instances of the custom data generators for training and validation data\n",
    "training_generator = DataGenerator(input_files_train)\n",
    "validation_generator = ValDataGenerator(input_files_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bfc5df10-d63f-4245-bc34-04e8030c98ee",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f0f4393e-3574-421a-8222-e75f8c4ed8c7",
   "metadata": {},
   "source": [
    "## <span style=\"color:green;\">3. Model</span>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fb91699d-1d6c-4cf4-9add-07e36d55296f",
   "metadata": {},
   "source": [
    "#### *Then we come to the architectural design of our model.  <span style=\"color:red;\">Deeplab V3</span> architecture is used in this model and some changes have been applied to it.*"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cd14e48b-3196-451f-b433-0610f5709adf",
   "metadata": {},
   "source": [
    "#### *Since the data used in this project is a time series. Time distributed convolutional layers are used so that the model has the ability to understand time and the extracted features are also in terms of time. We also added a regularization kernel to adding penalty factors to the network layers to alter the weight propagation through the layers which facilitate the model to converge optimally. In addition, max pooling and <span style=\"color:red;\">CNN</span> layers with different dilation_rate are used to extract features with different and better dimensions and to reduce the amount of data*."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "851a166d-dd0f-4f96-933f-fcfc84e5dec5",
   "metadata": {},
   "source": [
    "#### *After the CNN layers and their normalization by Batch Normalization, it is the turn of the <span style=\"color:red;\">GRU</span> layers to form the hybrid architecture we need. These layers, like RNN layers, can model the time step. Finally, by adding a few Dense and dropout layers to avoid overfitting the model, we complete the computational part of the model and add the output layer with 11 neurons to recognize the 11 classes in the land cover rasters in binary form.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "cd4e9021-bf97-445f-97f3-33aebc616537",
   "metadata": {},
   "outputs": [],
   "source": [
    "regularizer=keras.regularizers.l2(1e-2)\n",
    "ALPHA = 0.8\n",
    "BETA = 0.2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "352b5802-edad-4f93-93e1-23b94a033914",
   "metadata": {},
   "outputs": [],
   "source": [
    "def CNN_GRU_Model(input_shape):\n",
    "    \"\"\"\n",
    "    Creates a CNN-GRU model for spatiotemporal data processing.\n",
    "\n",
    "    Parameters:\n",
    "    input_shape (tuple): Shape of the input data (time_steps, width, height, channels).\n",
    "\n",
    "    Returns:\n",
    "    model (tf.keras.Model): Compiled CNN-GRU model.\n",
    "    \"\"\"\n",
    "    \n",
    "    # Define the input layer with the given shape\n",
    "    inputs = tf.keras.Input(shape=input_shape) \n",
    "    \n",
    "    # First convolutional block with TimeDistributed layers\n",
    "    x = TimeDistributed(Conv2D(64, (3,3), activation='relu', kernel_regularizer=regularizer))(inputs) \n",
    "    x = TimeDistributed(MaxPooling2D(pool_size=(2, 2)))(x) \n",
    "    \n",
    "    # Second convolutional block\n",
    "    x = TimeDistributed(Conv2D(128, (3,3), activation='relu', kernel_regularizer=regularizer))(x)\n",
    "    x = TimeDistributed(MaxPooling2D(pool_size=(2, 2)))(x) \n",
    "\n",
    "    # Atrous (dilated) convolutions for multi-scale context aggregation\n",
    "    x2 = TimeDistributed(Conv2D(64, (3,3), dilation_rate=3, activation='relu', padding='same', kernel_regularizer=regularizer))(x)\n",
    "    x3 = TimeDistributed(Conv2D(64, (3,3), dilation_rate=6, activation='relu', padding='same', kernel_regularizer=regularizer))(x)    \n",
    "    x4 = TimeDistributed(Conv2D(64, (3,3), dilation_rate=12, activation='relu', padding='same', kernel_regularizer=regularizer))(x)\n",
    "    x5 = TimeDistributed(Conv2D(64, (3,3), dilation_rate=18, activation='relu', padding='same', kernel_regularizer=regularizer))(x)\n",
    "    x6 = TimeDistributed(Conv2D(64, (3,3), dilation_rate=24, activation='relu', padding='same', kernel_regularizer=regularizer))(x)  \n",
    "    \n",
    "    # Concatenate the outputs of the atrous convolutions\n",
    "    x6 = concatenate([x6, x5, x2, x3, x4])\n",
    "    \n",
    "    # Additional convolutional layers\n",
    "    x7 = TimeDistributed(Conv2D(32, (1,1), padding='same', kernel_regularizer=regularizer))(x6) \n",
    "    x = TimeDistributed(MaxPooling2D(pool_size=(2, 2)))(x7)\n",
    "    x = TimeDistributed(Conv2D(32, (3,3), activation='relu', kernel_regularizer=regularizer))(x) \n",
    "    x = TimeDistributed(Conv2D(32, (3,3), activation='relu', kernel_regularizer=regularizer))(x) \n",
    "    x = TimeDistributed(BatchNormalization())(x) \n",
    "    x = TimeDistributed(Flatten())(x) \n",
    "    \n",
    "    # GRU layers for temporal processing\n",
    "    x = GRU(64, return_sequences=True)(x) \n",
    "    x = GRU(64)(x) \n",
    "\n",
    "    # Fully connected layers with Dropout\n",
    "    x = Dense(128, activation='relu', kernel_regularizer=regularizer)(x)\n",
    "    x = Dropout(rate=0.3)(x)\n",
    "    x = Dense(64, activation='relu', kernel_regularizer=regularizer)(x) \n",
    "    x = Dropout(rate=0.3)(x)\n",
    "    \n",
    "    # Output layer with sigmoid activation and reshaping to the desired output shape\n",
    "    outputs = Dense(64*64*11, activation='sigmoid')(x) \n",
    "    outputs = Reshape((64, 64, 11))(outputs) \n",
    "\n",
    "    # Create the model\n",
    "    model = tf.keras.Model(inputs=inputs, outputs=outputs)\n",
    "\n",
    "    return model\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6b9e54e2-ace2-475a-a97e-c817484dc804",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "adda0117-b8c5-4cf9-952e-e33be2ca7b67",
   "metadata": {},
   "source": [
    "### *Here we create two functions to calculate the confusion matrix, TverskyLoss, jac_distance and F1 score to use them for loss functios and evaluate our model.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "51fde3aa-4700-4378-87dc-bb95ddc7431e",
   "metadata": {},
   "outputs": [],
   "source": [
    "def TverskyLoss(targets, inputs, alpha=ALPHA, beta=BETA, smooth=1e-6):\n",
    "    \"\"\"\n",
    "    Computes the Tversky loss for image segmentation tasks.\n",
    "\n",
    "    The Tversky loss is a generalization of the Dice coefficient, which allows for \n",
    "    asymmetric penalties for false positives and false negatives. This can be useful \n",
    "    in medical image segmentation where the cost of false positives and false negatives \n",
    "    may differ.\n",
    "\n",
    "    Parameters:\n",
    "    targets (tensor): Ground truth binary masks.\n",
    "    inputs (tensor): Predicted binary masks.\n",
    "    alpha (float): Weight of false positives.\n",
    "    beta (float): Weight of false negatives.\n",
    "    smooth (float): Smoothing factor to avoid division by zero.\n",
    "\n",
    "    Returns:\n",
    "    float: Tversky loss value.\n",
    "    \"\"\"\n",
    "    \n",
    "    # Flatten the input and target tensors\n",
    "    inputs = K.flatten(inputs)\n",
    "    targets = K.flatten(targets)\n",
    "    \n",
    "    # Calculate true positives, false positives, and false negatives\n",
    "    TP = K.sum((inputs * targets))\n",
    "    FP = K.sum(((1 - targets) * inputs))\n",
    "    FN = K.sum((targets * (1 - inputs)))\n",
    "    \n",
    "    # Compute the Tversky index\n",
    "    Tversky = (TP + smooth) / (TP + alpha * FP + beta * FN + smooth)\n",
    "\n",
    "    # Return the Tversky loss\n",
    "    return (1 - Tversky)\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "e04662f4-9df3-46eb-9567-cef7181263e0",
   "metadata": {},
   "outputs": [],
   "source": [
    "def iou(y_true, y_pred, smooth=1):\n",
    "    \"\"\"\n",
    "    Computes the Intersection over Union (IoU) metric.\n",
    "\n",
    "    The IoU metric, also known as the Jaccard index, is used to evaluate the \n",
    "    similarity between two sets, typically used for image segmentation tasks.\n",
    "\n",
    "    Parameters:\n",
    "    y_true (tensor): Ground truth binary masks.\n",
    "    y_pred (tensor): Predicted binary masks.\n",
    "    smooth (float): Smoothing factor to avoid division by zero.\n",
    "\n",
    "    Returns:\n",
    "    float: IoU value.\n",
    "    \"\"\"\n",
    "    # Calculate the intersection and union of the true and predicted masks\n",
    "    intersection = K.sum(y_true * y_pred)\n",
    "    sum_ = K.sum(y_true + y_pred)\n",
    "    \n",
    "    # Compute the IoU\n",
    "    jac = (intersection + smooth) / (sum_ - intersection + smooth)\n",
    "    \n",
    "    return jac\n",
    "\n",
    "def jac_distance(y_true, y_pred):\n",
    "    \"\"\"\n",
    "    Computes the Jaccard distance, which is the complement of the IoU.\n",
    "\n",
    "    The Jaccard distance is used as a loss function for training segmentation models.\n",
    "\n",
    "    Parameters:\n",
    "    y_true (tensor): Ground truth binary masks.\n",
    "    y_pred (tensor): Predicted binary masks.\n",
    "\n",
    "    Returns:\n",
    "    float: Jaccard distance value.\n",
    "    \"\"\"\n",
    "    # Flatten the input tensors\n",
    "    y_truef = K.flatten(y_true)\n",
    "    y_predf = K.flatten(y_pred)\n",
    "    \n",
    "    # Return the negative IoU as the Jaccard distance\n",
    "    return -iou(y_true, y_pred)\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "676dec3e-b149-4e82-afc9-546182a3bf55",
   "metadata": {},
   "outputs": [],
   "source": [
    "def confusion_matrix(label, predict):\n",
    "    \"\"\"\n",
    "    Computes the confusion matrix for binary classification tasks.\n",
    "\n",
    "    The confusion matrix is a table that is often used to describe the performance \n",
    "    of a classification model. It contains the counts of true positives, true negatives, \n",
    "    false positives, and false negatives.\n",
    "\n",
    "    Parameters:\n",
    "    label (tensor): Ground truth binary labels.\n",
    "    predict (tensor): Predicted binary labels.\n",
    "\n",
    "    Returns:\n",
    "    np.array: 2x2 confusion matrix.\n",
    "    \"\"\"\n",
    "    \n",
    "    # Reshape the labels and predictions to 1D tensors\n",
    "    label = tf.reshape(label, [-1])\n",
    "    predict = tf.reshape(predict, [-1])\n",
    "\n",
    "    # Cast the labels and predictions to boolean\n",
    "    label = tf.cast(label, tf.bool)\n",
    "    predict = tf.cast(predict, tf.bool)\n",
    "    \n",
    "    # Calculate true positives, true negatives, false positives, and false negatives\n",
    "    true_positive = tf.reduce_sum(tf.cast(tf.logical_and(label, predict), tf.int32))\n",
    "    true_negative = tf.reduce_sum(tf.cast(tf.logical_and(tf.logical_not(label), tf.logical_not(predict)), tf.int32))\n",
    "    false_positive = tf.reduce_sum(tf.cast(tf.logical_and(tf.logical_not(label), predict), tf.int32))\n",
    "    false_negative = tf.reduce_sum(tf.cast(tf.logical_and(label, tf.logical_not(predict)), tf.int32))\n",
    "    \n",
    "    # Return the confusion matrix as a numpy array\n",
    "    return np.array([[true_positive, false_positive], [false_negative, true_negative]])\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "bcddab81-1137-4b92-ba67-be1c4d2dfbdc",
   "metadata": {},
   "outputs": [],
   "source": [
    "def F1_score(y_true, y_pred):\n",
    "    \"\"\"\n",
    "    Computes the F1 score for binary classification tasks.\n",
    "\n",
    "    The F1 score is the harmonic mean of precision and recall, providing a balance \n",
    "    between the two. It is especially useful for imbalanced datasets.\n",
    "\n",
    "    Parameters:\n",
    "    y_true (tensor): Ground truth binary labels.\n",
    "    y_pred (tensor): Predicted binary labels.\n",
    "\n",
    "    Returns:\n",
    "    float: F1 score value.\n",
    "    \"\"\"\n",
    "    \n",
    "    # Calculate precision and recall\n",
    "    precision = Precision()(y_true, y_pred)\n",
    "    recall = Recall()(y_true, y_pred)\n",
    "    \n",
    "    # Compute the F1 score\n",
    "    f1_score = 2 * (precision * recall) / (precision + recall + tf.keras.backend.epsilon())\n",
    "\n",
    "    return f1_score\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1994d8e1-a2fa-4526-8f5a-96b8fdaa14b8",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "35c90a32-a924-410b-8096-4c9dd4cb9203",
   "metadata": {},
   "source": [
    "### *Next, we can create an instance of the custom neural network model with the specified input shape.Then we compile the model using the precision and recall criteria and the <span style=\"color:red;\">binary_crossentropy</span> loss function. We also add 2 callbacks to save the best model.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "aa793695-11cf-4ab3-ac5d-94405fc71e69",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_1 (InputLayer)            [(None, 4, 64, 64, 1 0                                            \n",
      "__________________________________________________________________________________________________\n",
      "time_distributed (TimeDistribut (None, 4, 62, 62, 12 13952       input_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "time_distributed_1 (TimeDistrib (None, 4, 31, 31, 12 0           time_distributed[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "time_distributed_2 (TimeDistrib (None, 4, 29, 29, 25 295168      time_distributed_1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "time_distributed_3 (TimeDistrib (None, 4, 14, 14, 25 0           time_distributed_2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "time_distributed_8 (TimeDistrib (None, 4, 14, 14, 64 147520      time_distributed_3[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "time_distributed_7 (TimeDistrib (None, 4, 14, 14, 64 147520      time_distributed_3[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "time_distributed_4 (TimeDistrib (None, 4, 14, 14, 64 147520      time_distributed_3[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "time_distributed_5 (TimeDistrib (None, 4, 14, 14, 64 147520      time_distributed_3[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "time_distributed_6 (TimeDistrib (None, 4, 14, 14, 64 147520      time_distributed_3[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "concatenate (Concatenate)       (None, 4, 14, 14, 32 0           time_distributed_8[0][0]         \n",
      "                                                                 time_distributed_7[0][0]         \n",
      "                                                                 time_distributed_4[0][0]         \n",
      "                                                                 time_distributed_5[0][0]         \n",
      "                                                                 time_distributed_6[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "time_distributed_9 (TimeDistrib (None, 4, 14, 14, 32 10272       concatenate[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "time_distributed_10 (TimeDistri (None, 4, 7, 7, 32)  0           time_distributed_9[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "time_distributed_11 (TimeDistri (None, 4, 5, 5, 32)  9248        time_distributed_10[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "time_distributed_12 (TimeDistri (None, 4, 3, 3, 32)  9248        time_distributed_11[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "time_distributed_13 (TimeDistri (None, 4, 3, 3, 32)  128         time_distributed_12[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "time_distributed_14 (TimeDistri (None, 4, 288)       0           time_distributed_13[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "gru (GRU)                       (None, 4, 64)        67968       time_distributed_14[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "gru_1 (GRU)                     (None, 64)           24960       gru[0][0]                        \n",
      "__________________________________________________________________________________________________\n",
      "dense (Dense)                   (None, 128)          8320        gru_1[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "dropout (Dropout)               (None, 128)          0           dense[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "dense_1 (Dense)                 (None, 64)           8256        dropout[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dropout_1 (Dropout)             (None, 64)           0           dense_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dense_2 (Dense)                 (None, 32)           2080        dropout_1[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dense_3 (Dense)                 (None, 45056)        1486848     dense_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "reshape (Reshape)               (None, 64, 64, 11)   0           dense_3[0][0]                    \n",
      "==================================================================================================\n",
      "Total params: 2,674,048\n",
      "Trainable params: 2,673,984\n",
      "Non-trainable params: 64\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# Define the model with the given input shape\n",
    "model = CNN_GRU_Model((time_step, width, height, channel))\n",
    "\n",
    "# Define the Hyperparameters\n",
    "optimizer = tf.keras.optimizers.Adam(learning_rate = 0.0005)\n",
    "model.compile(loss = TverskyLoss, optimizer = optimizer, metrics = [Recall(), Precision(), jac_distance])\n",
    "early_stopping = keras.callbacks.EarlyStopping(monitor = \"val_loss\", patience = 25)\n",
    "\n",
    "callbacks = [ModelCheckpoint(\"D:/Learning/GeoAi/Week8/Project/BestModel.h5\",\n",
    "                             verbose = 1, save_best_only = True), early_stopping]\n",
    "\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ddc89635-c3b9-496f-bb5d-156f57090a04",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4506763f-ef22-4dd1-bc95-03984e65b6e0",
   "metadata": {},
   "source": [
    "### *Now we fit the created model on our training data with 20 epochs and after finishing the training process of the model, we save the best epoch.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 156,
   "id": "52d15270-27c1-473e-9b72-3f91faf3e563",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n",
      "8750/8750 [==============================] - 829s 95ms/step - loss: 0.6221 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2467 - val_loss: 0.5485 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3059\n",
      "\n",
      "Epoch 00001: val_loss did not improve from 0.54849\n",
      "Epoch 2/20\n",
      "8750/8750 [==============================] - 810s 93ms/step - loss: 0.6221 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2463 - val_loss: 0.5485 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3052\n",
      "\n",
      "Epoch 00002: val_loss improved from 0.54849 to 0.54848, saving model to D:/Learning/GeoAi/Week8/Project\\BestModel.h5\n",
      "Epoch 3/20\n",
      "8750/8750 [==============================] - 797s 91ms/step - loss: 0.6221 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2465 - val_loss: 0.5485 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3071\n",
      "\n",
      "Epoch 00003: val_loss did not improve from 0.54848\n",
      "Epoch 4/20\n",
      "8750/8750 [==============================] - 825s 94ms/step - loss: 0.6222 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2464 - val_loss: 0.5485 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3052\n",
      "\n",
      "Epoch 00004: val_loss did not improve from 0.54848\n",
      "Epoch 5/20\n",
      "8750/8750 [==============================] - 835s 95ms/step - loss: 0.6221 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2464 - val_loss: 0.5485 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3064\n",
      "\n",
      "Epoch 00005: val_loss improved from 0.54848 to 0.54846, saving model to D:/Learning/GeoAi/Week8/Project\\BestModel.h5\n",
      "Epoch 6/20\n",
      "8750/8750 [==============================] - 833s 95ms/step - loss: 0.6222 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2467 - val_loss: 0.5485 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3060\n",
      "\n",
      "Epoch 00006: val_loss did not improve from 0.54846\n",
      "Epoch 7/20\n",
      "8750/8750 [==============================] - 831s 95ms/step - loss: 0.6221 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2467 - val_loss: 0.5485 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3052\n",
      "\n",
      "Epoch 00007: val_loss did not improve from 0.54846\n",
      "Epoch 8/20\n",
      "8750/8750 [==============================] - 830s 95ms/step - loss: 0.6222 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2465 - val_loss: 0.5485 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3062\n",
      "\n",
      "Epoch 00008: val_loss did not improve from 0.54846\n",
      "Epoch 9/20\n",
      "8750/8750 [==============================] - 830s 95ms/step - loss: 0.6221 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2463 - val_loss: 0.5485 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3058\n",
      "\n",
      "Epoch 00009: val_loss did not improve from 0.54846\n",
      "Epoch 10/20\n",
      "8750/8750 [==============================] - 822s 94ms/step - loss: 0.6222 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2466 - val_loss: 0.5486 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3061\n",
      "\n",
      "Epoch 00010: val_loss did not improve from 0.54846\n",
      "Epoch 11/20\n",
      "8750/8750 [==============================] - 830s 95ms/step - loss: 0.6221 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2466 - val_loss: 0.5485 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3058\n",
      "\n",
      "Epoch 00011: val_loss did not improve from 0.54846\n",
      "Epoch 12/20\n",
      "8750/8750 [==============================] - 832s 95ms/step - loss: 0.6221 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2466 - val_loss: 0.5486 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3055\n",
      "\n",
      "Epoch 00012: val_loss did not improve from 0.54846\n",
      "Epoch 13/20\n",
      "8750/8750 [==============================] - 831s 95ms/step - loss: 0.6221 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2466 - val_loss: 0.5486 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3059\n",
      "\n",
      "Epoch 00013: val_loss did not improve from 0.54846\n",
      "Epoch 14/20\n",
      "8750/8750 [==============================] - 831s 95ms/step - loss: 0.6221 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2467 - val_loss: 0.5484 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3068\n",
      "\n",
      "Epoch 00014: val_loss improved from 0.54846 to 0.54843, saving model to D:/Learning/GeoAi/Week8/Project\\BestModel.h5\n",
      "Epoch 15/20\n",
      "8750/8750 [==============================] - 830s 95ms/step - loss: 0.6222 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2467 - val_loss: 0.5483 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3059\n",
      "\n",
      "Epoch 00015: val_loss improved from 0.54843 to 0.54828, saving model to D:/Learning/GeoAi/Week8/Project\\BestModel.h5\n",
      "Epoch 16/20\n",
      "8750/8750 [==============================] - 831s 95ms/step - loss: 0.6221 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2465 - val_loss: 0.5485 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3060\n",
      "\n",
      "Epoch 00016: val_loss did not improve from 0.54828\n",
      "Epoch 17/20\n",
      "8750/8750 [==============================] - 831s 95ms/step - loss: 0.6221 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2465 - val_loss: 0.5486 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3063\n",
      "\n",
      "Epoch 00017: val_loss did not improve from 0.54828\n",
      "Epoch 18/20\n",
      "8750/8750 [==============================] - 829s 95ms/step - loss: 0.6222 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2464 - val_loss: 0.5485 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3052\n",
      "\n",
      "Epoch 00018: val_loss did not improve from 0.54828\n",
      "Epoch 19/20\n",
      "8750/8750 [==============================] - 827s 94ms/step - loss: 0.6221 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2467 - val_loss: 0.5484 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3061\n",
      "\n",
      "Epoch 00019: val_loss did not improve from 0.54828\n",
      "Epoch 20/20\n",
      "8750/8750 [==============================] - 827s 94ms/step - loss: 0.6221 - recall_6: 0.3886 - precision_6: 0.3758 - jac_distance: -0.2465 - val_loss: 0.5484 - val_recall_6: 0.4614 - val_precision_6: 0.4497 - val_jac_distance: -0.3064\n",
      "\n",
      "Epoch 00020: val_loss did not improve from 0.54828\n"
     ]
    }
   ],
   "source": [
    "history = model.fit(training_generator, epochs = 20, validation_data = validation_generator, callbacks = callbacks)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fdb108ef-c147-4182-ba73-c9d117928bda",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3e3ebf90-b155-4f94-b3b3-cfd71f2a3338",
   "metadata": {},
   "source": [
    "## <span style=\"color:green;\">4. Evaluation</span>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "79f0e4d0-2927-4a5f-aba2-c0e957e92036",
   "metadata": {},
   "source": [
    "### *Next, we load the best model for predict the test data and evaluate it based on the mentioned criteria.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "4ddfde96-db5a-4a52-ae2c-0e3354d78a8b",
   "metadata": {},
   "outputs": [],
   "source": [
    "MyModel = load_model(\"D:\\Learning\\GeoAi\\Week8\\Project\\BestModel.h5\", custom_objects = {'TverskyLoss':TverskyLoss, 'jac_distance':jac_distance})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "id": "258d7bba-d257-4b60-bec4-9226c51d5067",
   "metadata": {},
   "outputs": [],
   "source": [
    "label_pred = MyModel.predict(validation_generator)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "339b022c-dc80-4c1a-b6e6-7838cfad8e53",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "625/625 [==============================] - 11s 15ms/step - loss: 0.5485 - recall_6: 0.4614 - precision_6: 0.4497 - jac_distance: -0.3056\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.5485276579856873,\n",
       " 0.4614284634590149,\n",
       " 0.44965338706970215,\n",
       " -0.3055548071861267]"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "MyModel.evaluate(validation_generator)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "acd03e5a-c1b2-41cb-8dd8-71de02d8e466",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "32563565-9d52-41cb-9a0a-bc5927f06e91",
   "metadata": {},
   "source": [
    "### *In the following, the <span style=\"color:red;\">sigmoid</span> activator outputs the values as numbers between 0 and 1. We create a threshold on the output data so that each element becomes 0 or 1.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "e2bbbd54-6f2c-4f2f-b7a3-b345085be786",
   "metadata": {},
   "outputs": [],
   "source": [
    "condition = label_pred > 0.5\n",
    "label_pred = np.where(condition, 1, 0)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cf556eaf-6268-4b02-8717-5d9376ccfc43",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "027edbc5-87e2-4c73-bb60-f96de3f5c29f",
   "metadata": {},
   "source": [
    "### *Now we create a matrix to cover all the project data using the stored arrays.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "id": "f6c1612f-69ee-45c7-96a7-e1fa9a17f595",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Initialize lists to store input and label arrays\n",
    "input_list = []\n",
    "label_list = []\n",
    "\n",
    "# Get the number of files in the input directory\n",
    "num_files = len(os.listdir(\"D:\\\\Learning\\\\GeoAi\\\\Week8\\\\Project\\\\Test\\\\Input\"))\n",
    "\n",
    "# Loop through each file index\n",
    "for i in range(num_files):\n",
    "   \n",
    "    input_file = \"D:\\\\Learning\\\\GeoAi\\\\Week8\\\\Project\\\\Test\\\\Input\\\\input_\" + str(i) + '.npz'\n",
    "    label_file = \"D:\\\\Learning\\\\GeoAi\\\\Week8\\\\Project\\\\Test\\\\Label\\\\label_\" + str(i) + '.npz'\n",
    "    \n",
    "    # Load the input and label arrays from the .npz files\n",
    "    input_array = np.load(input_file)['input_train']\n",
    "    label_array = np.load(label_file)['label_train']\n",
    "    \n",
    "    # Append the arrays to the respective lists\n",
    "    input_list.append(input_array)\n",
    "    label_list.append(label_array)\n",
    "\n",
    "# Stack the lists into 5D and 4D arrays respectively\n",
    "input_5D = np.stack(input_list, axis=0)\n",
    "label_4D = np.stack(label_list, axis=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cc24860e-c121-4d67-9ff5-fd9abff168da",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6590b5f1-88bd-422e-b5ab-5c5ea5cfd321",
   "metadata": {},
   "source": [
    "### *Then we come to evaluations of confusion matrix and F1 score.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "b6c0ffc5-9ceb-403e-8744-37d9bcdda160",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Positive</th>\n",
       "      <th>Negative</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>True</th>\n",
       "      <td>4614568.0</td>\n",
       "      <td>5647932.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>False</th>\n",
       "      <td>5386046.0</td>\n",
       "      <td>96991454.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        Positive    Negative\n",
       "True   4614568.0   5647932.0\n",
       "False  5386046.0  96991454.0"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "total_confusion_matrix = np.zeros((2, 2))\n",
    "\n",
    "for i in range(2500):\n",
    "    total_confusion_matrix += confusion_matrix(label_4D[i], label_pred[i])\n",
    "\n",
    "total_confusion_matrix = pd.DataFrame(total_confusion_matrix, index=['True', 'False'], columns=['Positive', 'Negative'])\n",
    "total_confusion_matrix"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e56ee8d8-ca6a-4c28-8607-582fa9608045",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "id": "2f084d69-fa24-4011-85b7-c3037aae6056",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 800x600 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot the dataframe as a heatmap using seaborn\n",
    "plt.figure(figsize=(8, 6))\n",
    "sb.heatmap(total_confusion_matrix, annot=True, fmt='.0f', cmap=\"BuPu\")\n",
    "plt.title('Confusion Matrix')\n",
    "plt.xlabel('Predicted')\n",
    "plt.ylabel('Actual')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "554afeee-95e4-4718-be49-afb13bda0e7e",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e89eec91-c5fb-44f7-a227-a2386c5fc819",
   "metadata": {},
   "outputs": [],
   "source": [
    "f1_scores = []\n",
    "\n",
    "for i in range(2500):\n",
    "    label = label_4D[i].flatten()\n",
    "    predict = label_pred[i].flatten()\n",
    "    f1 = f1_score(label, predict)\n",
    "    f1_scores.append(f1)\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "id": "276f562e-0786-4064-9700-183a07ebd24d",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 Score :  0.4502526237820023\n"
     ]
    }
   ],
   "source": [
    "mean_f1_score = (sum(f1_scores)/len(f1_scores))\n",
    "print('F1 Score : ', mean_f1_score)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a536a14f-4d09-40f2-aa0a-a55b85d49471",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "47d05e89-9964-4cc3-8bbc-14d1dae76844",
   "metadata": {},
   "source": [
    "### *Now, we randomly select 5 patches for decoding to visually compare the predicted map with the initial map.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 176,
   "id": "1887467b-748b-4f97-8891-de68215b085b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1200x1000 with 10 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Define the class values for the land cover types\n",
    "val = [20, 31, 32, 33, 50, 80, 81, 100, 210, 220, 230]\n",
    "\n",
    "# Initialize lists to store the final predicted and label maps\n",
    "final_pred = []\n",
    "final_label = []\n",
    "\n",
    "# Loop to generate and visualize 5 random samples\n",
    "for j in range(5):\n",
    "    \n",
    "    # Initialize empty maps for predictions and labels\n",
    "    final_map_pred = np.zeros((64, 64), dtype=np.float32)\n",
    "    final_map_label = np.zeros((64, 64), dtype=np.float32)\n",
    "    \n",
    "    # Select a random sample index\n",
    "    k = np.random.randint(0, 2499)\n",
    "    print(k)\n",
    "    \n",
    "    # Map the predicted and label values to the final maps\n",
    "    for i in range(11):\n",
    "        x, y = np.where(label_pred[k, :, :, i] == 1)\n",
    "        final_map_pred[x, y] = val[i]\n",
    "        \n",
    "        u, v = np.where(label_4D[k, :, :, i] == 1)\n",
    "        final_map_label[u, v] = val[i] \n",
    "        \n",
    "    # Append the final maps to the lists\n",
    "    final_pred.append(final_map_pred)\n",
    "    final_label.append(final_map_label)\n",
    "\n",
    "fig, axes = plt.subplots(2, 5, figsize=(6, 5), dpi=200)\n",
    "\n",
    "# Set titles for the label and prediction rows\n",
    "axes[0, 2].set_title(\"Label Patches\", fontsize=16, y=1.1)\n",
    "axes[1, 2].set_title(\"Predict Patches\", fontsize=16, y=1.1)\n",
    "\n",
    "# Loop to display the label and prediction maps\n",
    "for i in range(5):\n",
    "    axes[0, i].imshow(final_label[i], cmap='viridis')\n",
    "    axes[0, i].set_xticks([])\n",
    "    axes[0, i].set_yticks([])\n",
    "    \n",
    "    axes[1, i].imshow(final_pred[i], cmap='viridis')\n",
    "    axes[1, i].set_xticks([])\n",
    "    axes[1, i].set_yticks([])\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "942c2cb1-767f-4f3e-aa4f-daf701c7d928",
   "metadata": {
    "id": "zAwRP94YJsrb"
   },
   "source": [
    "-------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4ae14746-9583-461e-9b08-a8fe3e13dc7a",
   "metadata": {},
   "source": [
    "### *As we can see, by examining the random patches, we find that the model has made most of the patches belong to the class with a value 210. This problem of low accuracy of the model can be due to the complexity of the data pattern, which makes the model with 20 epochs not understand it. Therefore, to improve the accuracy of this model, the number of epochs can be increased, or more data can be used to train the model. Also, the selection of the Loss function can have a favorable effect on the prediction result.*"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9e152d7e-8a7a-488c-8630-9591783d83be",
   "metadata": {},
   "source": [
    "# END ^^"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "gpu-env",
   "language": "python",
   "name": "gpu-env"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.18"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}