Classification and detection of cars in Drone images¶
We start by installing and importing the necessary libraries:
In [ ]:
!pip install rasterio
Collecting rasterio Downloading rasterio-1.4.3-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (9.1 kB) Collecting affine (from rasterio) Downloading affine-2.4.0-py3-none-any.whl.metadata (4.0 kB) Requirement already satisfied: attrs in /usr/local/lib/python3.10/dist-packages (from rasterio) (24.3.0) Requirement already satisfied: certifi in /usr/local/lib/python3.10/dist-packages (from rasterio) (2024.12.14) Requirement already satisfied: click>=4.0 in /usr/local/lib/python3.10/dist-packages (from rasterio) (8.1.8) Collecting cligj>=0.5 (from rasterio) Downloading cligj-0.7.2-py3-none-any.whl.metadata (5.0 kB) Requirement already satisfied: numpy>=1.24 in /usr/local/lib/python3.10/dist-packages (from rasterio) (1.26.4) Collecting click-plugins (from rasterio) Downloading click_plugins-1.1.1-py2.py3-none-any.whl.metadata (6.4 kB) Requirement already satisfied: pyparsing in /usr/local/lib/python3.10/dist-packages (from rasterio) (3.2.1) Downloading rasterio-1.4.3-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (22.2 MB) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 22.2/22.2 MB 46.2 MB/s eta 0:00:00 Downloading cligj-0.7.2-py3-none-any.whl (7.1 kB) Downloading affine-2.4.0-py3-none-any.whl (15 kB) Downloading click_plugins-1.1.1-py2.py3-none-any.whl (7.5 kB) Installing collected packages: cligj, click-plugins, affine, rasterio Successfully installed affine-2.4.0 click-plugins-1.1.1 cligj-0.7.2 rasterio-1.4.3
In [ ]:
from google.colab import drive
drive.mount('/content/drive')
Mounted at /content/drive
In [ ]:
import rasterio
import numpy as np
from matplotlib import pyplot as plt
import cv2
from shapely.geometry import shape
import geopandas as gpd
import shapely
from rasterio.plot import show
from sklearn.model_selection import train_test_split
import pandas as pd
We define the image path and import it, converting it into a numpy array, to then plot:
In [ ]:
path = 'drive/My Drive/Datasets/Car_detection/L1_car_detect.tif'
In [ ]:
src = rasterio.open(path)
im = src.read()
In [ ]:
im = im.transpose([1,2,0])
In [ ]:
im.shape
Out[ ]:
(12715, 12104, 4)
In [ ]:
im = im.astype('uint8')
In [ ]:
plt.figure(figsize=[16,16])
plt.imshow(im)
plt.axis('off')
Out[ ]:
(-0.5, 12103.5, 12714.5, -0.5)
Let's now import our reference data. The Drone image was opened in Qgis and two shapefiles were created. One with car location points to create the positive samples and the other with location points of other objects in the image to create the negative samples. Let's import these two files and plot the points on the image. In red the cars and in blue non-cars.
In [ ]:
Cars = gpd.read_file('drive/My Drive/Datasets/Car_detection/Cars.shp')
No_Cars = gpd.read_file('drive/My Drive/Datasets/Car_detection/No_Cars.shp')
In [ ]:
fig, ax = plt.subplots(figsize=(20, 20))
show((src), ax=ax)
Cars.plot(ax=ax, marker='o', color='red', markersize=15)
No_Cars.plot(ax=ax, marker='o', color='blue', markersize=15)
Out[ ]:
<Axes: >
The next step is to extract images from each point. Using rasterio and numpy, we will create a 40-pixel buffer around each point and cut it, making the resulting image 80x80x3 in size. We will also apply some data augmentation operations like rotation and flips.
In [ ]:
Car_img_list = []
for car_point in Cars['geometry']:
x = car_point.xy[0][0]
y = car_point.xy[1][0]
row, col = src.index(x,y)
img_path = im[row-40:row+40,col-40:col+40,0:3]
flipVertical = cv2.flip(img_path, 0)
flipHorizontal = cv2.flip(img_path, 1)
flipBoth = cv2.flip(img_path, -1)
image90 = cv2.rotate(img_path, cv2.ROTATE_90_CLOCKWISE)
rows,cols,_ = img_path.shape
M = cv2.getRotationMatrix2D(((cols-1)/2.0,(rows-1)/2.0),45,1)
dst = cv2.warpAffine(img_path,M,(cols,rows))
Car_img_list.append(img_path)
Car_img_list.append(flipVertical)
Car_img_list.append(flipHorizontal)
Car_img_list.append(flipBoth)
Car_img_list.append(image90)
Car_img_list.append(dst)
Car_img_list = np.array(Car_img_list)
In [ ]:
No_Car_img_list = []
for No_car_point in No_Cars['geometry']:
x = No_car_point.xy[0][0]
y = No_car_point.xy[1][0]
row, col = src.index(x,y)
img_path = im[row-40:row+40,col-40:col+40,0:3]
flipVertical = cv2.flip(img_path, 0)
flipHorizontal = cv2.flip(img_path, 1)
flipBoth = cv2.flip(img_path, -1)
image90 = cv2.rotate(img_path, cv2.ROTATE_90_CLOCKWISE)
rows,cols,_ = img_path.shape
M = cv2.getRotationMatrix2D(((cols-1)/2.0,(rows-1)/2.0),45,1)
dst = cv2.warpAffine(img_path,M,(cols,rows))
No_Car_img_list.append(img_path)
No_Car_img_list.append(flipVertical)
No_Car_img_list.append(flipHorizontal)
No_Car_img_list.append(flipBoth)
No_Car_img_list.append(image90)
No_Car_img_list.append(dst)
No_Car_img_list = np.array(No_Car_img_list)
In [ ]:
Car_img_list.shape
Out[ ]:
(852, 80, 80, 3)
Let's plot an example of a car image and a non-car image:
In [ ]:
plt.figure(figsize=[4,4])
plt.imshow(Car_img_list[6])
Out[ ]:
<matplotlib.image.AxesImage at 0x7f8aca8de6d0>
In [ ]:
plt.figure(figsize=[4,4])
plt.imshow(No_Car_img_list[0])
Out[ ]:
<matplotlib.image.AxesImage at 0x7f8aca8a4250>
We can now create reference Y vectors with values of 0 for non-car and 1 for car. Then we normalize the values of the image dataset to the range of 0 to 1.
In [ ]:
Y_car = np.ones(len(Car_img_list))
Y_No_car = np.zeros(len(No_Car_img_list))
In [ ]:
X_car = np.array(Car_img_list)
X_No_car = np.array(No_Car_img_list)
In [ ]:
X = np.concatenate((X_car,X_No_car), axis=0)
Y = np.concatenate((Y_car,Y_No_car), axis=0)
In [ ]:
X = X/255
The next step is to divide the data into training and testing.
In [ ]:
x_train, x_test, y_train, y_test = train_test_split(X, Y, test_size = 0.3, random_state = 42)
And then the creation of the custom convolutional network for image classification:
In [ ]:
from keras.models import Model, Sequential
from keras.layers import Input, Conv2D, MaxPooling2D, Flatten, Dense, BatchNormalization, Dropout, Activation
#from tensorflow.keras.optimizers import Adam
from keras.optimizers import Adam
from keras.activations import relu
from keras.losses import binary_crossentropy
from keras import backend as K
import tensorflow as tf
from keras.models import load_model
In [ ]:
Model = Sequential()
Model.add(Conv2D(32,(3,3), kernel_initializer="he_normal", padding='same', input_shape=(x_train.shape[1:])))
Model.add(Activation('relu'))
Model.add(MaxPooling2D(pool_size=(2,2),strides=2))
Model.add(BatchNormalization())
Model.add(Dropout(0.25))
Model.add(Conv2D(64,(3,3), kernel_initializer="he_normal", padding='same'))
Model.add(Activation('relu'))
Model.add(MaxPooling2D(pool_size=(2,2),strides=2))
Model.add(BatchNormalization())
Model.add(Dropout(0.25))
Model.add(Conv2D(128,(3,3), kernel_initializer="he_normal", padding='same'))
Model.add(Activation('relu'))
Model.add(MaxPooling2D(pool_size=(2,2),strides=2))
Model.add(Flatten())
Model.add(Dense(128))
Model.add(Activation('relu'))
Model.add(Dropout(0.25))
Model.add(Dense(32))
Model.add(Activation('relu'))
Model.add(Dense(1))
Model.add(Activation('sigmoid'))
Model.compile(loss='binary_crossentropy', optimizer=Adam(learning_rate=0.00001), metrics=['accuracy'])
Model.summary()
Model: "sequential_2"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d_3 (Conv2D) (None, 80, 80, 32) 896
activation_6 (Activation) (None, 80, 80, 32) 0
max_pooling2d_3 (MaxPooling (None, 40, 40, 32) 0
2D)
batch_normalization_2 (Batc (None, 40, 40, 32) 128
hNormalization)
dropout_3 (Dropout) (None, 40, 40, 32) 0
conv2d_4 (Conv2D) (None, 40, 40, 64) 18496
activation_7 (Activation) (None, 40, 40, 64) 0
max_pooling2d_4 (MaxPooling (None, 20, 20, 64) 0
2D)
batch_normalization_3 (Batc (None, 20, 20, 64) 256
hNormalization)
dropout_4 (Dropout) (None, 20, 20, 64) 0
conv2d_5 (Conv2D) (None, 20, 20, 128) 73856
activation_8 (Activation) (None, 20, 20, 128) 0
max_pooling2d_5 (MaxPooling (None, 10, 10, 128) 0
2D)
flatten_1 (Flatten) (None, 12800) 0
dense_3 (Dense) (None, 128) 1638528
activation_9 (Activation) (None, 128) 0
dropout_5 (Dropout) (None, 128) 0
dense_4 (Dense) (None, 32) 4128
activation_10 (Activation) (None, 32) 0
dense_5 (Dense) (None, 1) 33
activation_11 (Activation) (None, 1) 0
=================================================================
Total params: 1,736,321
Trainable params: 1,736,129
Non-trainable params: 192
_________________________________________________________________
/usr/local/lib/python3.7/dist-packages/keras/optimizer_v2/adam.py:105: UserWarning: The `lr` argument is deprecated, use `learning_rate` instead. super(Adam, self).__init__(name, **kwargs)
In [ ]:
history = Model.fit(x = x_train, y= y_train, batch_size=64, epochs=200, verbose=1, shuffle=True, validation_split=0.2)
Epoch 1/200 25/25 [==============================] - 1s 30ms/step - loss: 0.5920 - accuracy: 0.7151 - val_loss: 0.6846 - val_accuracy: 0.5882 Epoch 2/200 25/25 [==============================] - 1s 20ms/step - loss: 0.4541 - accuracy: 0.7926 - val_loss: 0.6622 - val_accuracy: 0.6113 Epoch 3/200 25/25 [==============================] - 1s 20ms/step - loss: 0.3972 - accuracy: 0.8175 - val_loss: 0.5847 - val_accuracy: 0.7954 Epoch 4/200 25/25 [==============================] - 1s 20ms/step - loss: 0.3518 - accuracy: 0.8406 - val_loss: 0.5416 - val_accuracy: 0.8312 Epoch 5/200 25/25 [==============================] - 1s 20ms/step - loss: 0.3271 - accuracy: 0.8502 - val_loss: 0.5232 - val_accuracy: 0.8261 Epoch 6/200 25/25 [==============================] - 1s 25ms/step - loss: 0.3069 - accuracy: 0.8758 - val_loss: 0.4722 - val_accuracy: 0.8593 Epoch 7/200 25/25 [==============================] - 1s 20ms/step - loss: 0.2779 - accuracy: 0.8835 - val_loss: 0.3842 - val_accuracy: 0.9003 Epoch 8/200 25/25 [==============================] - 1s 24ms/step - loss: 0.2774 - accuracy: 0.8828 - val_loss: 0.3565 - val_accuracy: 0.9054 Epoch 9/200 25/25 [==============================] - 0s 20ms/step - loss: 0.2742 - accuracy: 0.8873 - val_loss: 0.3255 - val_accuracy: 0.8875 Epoch 10/200 25/25 [==============================] - 0s 20ms/step - loss: 0.2486 - accuracy: 0.8988 - val_loss: 0.2899 - val_accuracy: 0.9079 Epoch 11/200 25/25 [==============================] - 1s 20ms/step - loss: 0.2314 - accuracy: 0.8982 - val_loss: 0.2894 - val_accuracy: 0.9105 Epoch 12/200 25/25 [==============================] - 1s 20ms/step - loss: 0.2263 - accuracy: 0.9027 - val_loss: 0.2528 - val_accuracy: 0.9130 Epoch 13/200 25/25 [==============================] - 0s 20ms/step - loss: 0.2056 - accuracy: 0.9181 - val_loss: 0.2422 - val_accuracy: 0.9003 Epoch 14/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1952 - accuracy: 0.9251 - val_loss: 0.2262 - val_accuracy: 0.9130 Epoch 15/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1828 - accuracy: 0.9353 - val_loss: 0.2187 - val_accuracy: 0.9130 Epoch 16/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1757 - accuracy: 0.9353 - val_loss: 0.2126 - val_accuracy: 0.9130 Epoch 17/200 25/25 [==============================] - 1s 20ms/step - loss: 0.1890 - accuracy: 0.9219 - val_loss: 0.2169 - val_accuracy: 0.9028 Epoch 18/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1643 - accuracy: 0.9385 - val_loss: 0.2059 - val_accuracy: 0.9105 Epoch 19/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1769 - accuracy: 0.9315 - val_loss: 0.1989 - val_accuracy: 0.9130 Epoch 20/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1676 - accuracy: 0.9366 - val_loss: 0.1867 - val_accuracy: 0.9233 Epoch 21/200 25/25 [==============================] - 1s 20ms/step - loss: 0.1629 - accuracy: 0.9385 - val_loss: 0.1844 - val_accuracy: 0.9258 Epoch 22/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1649 - accuracy: 0.9398 - val_loss: 0.1802 - val_accuracy: 0.9233 Epoch 23/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1567 - accuracy: 0.9385 - val_loss: 0.1964 - val_accuracy: 0.9028 Epoch 24/200 25/25 [==============================] - 1s 20ms/step - loss: 0.1448 - accuracy: 0.9392 - val_loss: 0.2197 - val_accuracy: 0.9054 Epoch 25/200 25/25 [==============================] - 1s 20ms/step - loss: 0.1537 - accuracy: 0.9430 - val_loss: 0.2002 - val_accuracy: 0.9105 Epoch 26/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1501 - accuracy: 0.9411 - val_loss: 0.1783 - val_accuracy: 0.9284 Epoch 27/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1321 - accuracy: 0.9545 - val_loss: 0.1838 - val_accuracy: 0.9233 Epoch 28/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1314 - accuracy: 0.9539 - val_loss: 0.1978 - val_accuracy: 0.9156 Epoch 29/200 25/25 [==============================] - 1s 20ms/step - loss: 0.1240 - accuracy: 0.9526 - val_loss: 0.1807 - val_accuracy: 0.9207 Epoch 30/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1341 - accuracy: 0.9494 - val_loss: 0.2021 - val_accuracy: 0.9105 Epoch 31/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1319 - accuracy: 0.9469 - val_loss: 0.2151 - val_accuracy: 0.8977 Epoch 32/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1162 - accuracy: 0.9533 - val_loss: 0.1729 - val_accuracy: 0.9233 Epoch 33/200 25/25 [==============================] - 1s 20ms/step - loss: 0.1250 - accuracy: 0.9526 - val_loss: 0.1830 - val_accuracy: 0.9182 Epoch 34/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1102 - accuracy: 0.9603 - val_loss: 0.1980 - val_accuracy: 0.9079 Epoch 35/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1156 - accuracy: 0.9513 - val_loss: 0.2156 - val_accuracy: 0.9028 Epoch 36/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1013 - accuracy: 0.9641 - val_loss: 0.1981 - val_accuracy: 0.9105 Epoch 37/200 25/25 [==============================] - 1s 20ms/step - loss: 0.1052 - accuracy: 0.9622 - val_loss: 0.1928 - val_accuracy: 0.9156 Epoch 38/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1058 - accuracy: 0.9609 - val_loss: 0.1908 - val_accuracy: 0.9182 Epoch 39/200 25/25 [==============================] - 0s 20ms/step - loss: 0.1012 - accuracy: 0.9616 - val_loss: 0.2025 - val_accuracy: 0.9079 Epoch 40/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0983 - accuracy: 0.9629 - val_loss: 0.2007 - val_accuracy: 0.9156 Epoch 41/200 25/25 [==============================] - 1s 20ms/step - loss: 0.1062 - accuracy: 0.9565 - val_loss: 0.2117 - val_accuracy: 0.9079 Epoch 42/200 25/25 [==============================] - 1s 20ms/step - loss: 0.1007 - accuracy: 0.9641 - val_loss: 0.1980 - val_accuracy: 0.9182 Epoch 43/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0945 - accuracy: 0.9693 - val_loss: 0.2432 - val_accuracy: 0.8926 Epoch 44/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0826 - accuracy: 0.9680 - val_loss: 0.2151 - val_accuracy: 0.9105 Epoch 45/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0846 - accuracy: 0.9680 - val_loss: 0.1924 - val_accuracy: 0.9182 Epoch 46/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0817 - accuracy: 0.9718 - val_loss: 0.1953 - val_accuracy: 0.9156 Epoch 47/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0805 - accuracy: 0.9712 - val_loss: 0.1921 - val_accuracy: 0.9130 Epoch 48/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0837 - accuracy: 0.9648 - val_loss: 0.1869 - val_accuracy: 0.9182 Epoch 49/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0936 - accuracy: 0.9603 - val_loss: 0.2425 - val_accuracy: 0.8977 Epoch 50/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0885 - accuracy: 0.9654 - val_loss: 0.2201 - val_accuracy: 0.9079 Epoch 51/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0758 - accuracy: 0.9744 - val_loss: 0.2111 - val_accuracy: 0.9105 Epoch 52/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0854 - accuracy: 0.9686 - val_loss: 0.2073 - val_accuracy: 0.9156 Epoch 53/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0753 - accuracy: 0.9770 - val_loss: 0.2158 - val_accuracy: 0.9130 Epoch 54/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0763 - accuracy: 0.9738 - val_loss: 0.2543 - val_accuracy: 0.8977 Epoch 55/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0684 - accuracy: 0.9757 - val_loss: 0.2042 - val_accuracy: 0.9156 Epoch 56/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0770 - accuracy: 0.9725 - val_loss: 0.2565 - val_accuracy: 0.8951 Epoch 57/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0717 - accuracy: 0.9776 - val_loss: 0.2422 - val_accuracy: 0.9028 Epoch 58/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0729 - accuracy: 0.9712 - val_loss: 0.2252 - val_accuracy: 0.9105 Epoch 59/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0564 - accuracy: 0.9827 - val_loss: 0.2235 - val_accuracy: 0.9054 Epoch 60/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0707 - accuracy: 0.9725 - val_loss: 0.1933 - val_accuracy: 0.9130 Epoch 61/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0731 - accuracy: 0.9750 - val_loss: 0.2055 - val_accuracy: 0.9105 Epoch 62/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0622 - accuracy: 0.9782 - val_loss: 0.1989 - val_accuracy: 0.9130 Epoch 63/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0590 - accuracy: 0.9846 - val_loss: 0.2084 - val_accuracy: 0.9130 Epoch 64/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0635 - accuracy: 0.9776 - val_loss: 0.2272 - val_accuracy: 0.9079 Epoch 65/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0596 - accuracy: 0.9827 - val_loss: 0.2162 - val_accuracy: 0.9079 Epoch 66/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0548 - accuracy: 0.9834 - val_loss: 0.2318 - val_accuracy: 0.9079 Epoch 67/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0492 - accuracy: 0.9853 - val_loss: 0.2037 - val_accuracy: 0.9079 Epoch 68/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0564 - accuracy: 0.9840 - val_loss: 0.2211 - val_accuracy: 0.9105 Epoch 69/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0551 - accuracy: 0.9808 - val_loss: 0.2462 - val_accuracy: 0.9003 Epoch 70/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0513 - accuracy: 0.9834 - val_loss: 0.2001 - val_accuracy: 0.9156 Epoch 71/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0570 - accuracy: 0.9808 - val_loss: 0.2372 - val_accuracy: 0.9105 Epoch 72/200 25/25 [==============================] - 1s 23ms/step - loss: 0.0512 - accuracy: 0.9840 - val_loss: 0.2305 - val_accuracy: 0.9079 Epoch 73/200 25/25 [==============================] - 1s 22ms/step - loss: 0.0462 - accuracy: 0.9878 - val_loss: 0.2630 - val_accuracy: 0.9028 Epoch 74/200 25/25 [==============================] - 1s 27ms/step - loss: 0.0487 - accuracy: 0.9834 - val_loss: 0.1874 - val_accuracy: 0.9182 Epoch 75/200 25/25 [==============================] - 1s 24ms/step - loss: 0.0555 - accuracy: 0.9802 - val_loss: 0.2030 - val_accuracy: 0.9105 Epoch 76/200 25/25 [==============================] - 1s 29ms/step - loss: 0.0613 - accuracy: 0.9763 - val_loss: 0.2060 - val_accuracy: 0.9156 Epoch 77/200 25/25 [==============================] - 1s 25ms/step - loss: 0.0502 - accuracy: 0.9827 - val_loss: 0.1912 - val_accuracy: 0.9182 Epoch 78/200 25/25 [==============================] - 1s 23ms/step - loss: 0.0493 - accuracy: 0.9840 - val_loss: 0.2417 - val_accuracy: 0.9079 Epoch 79/200 25/25 [==============================] - 1s 22ms/step - loss: 0.0423 - accuracy: 0.9878 - val_loss: 0.1987 - val_accuracy: 0.9182 Epoch 80/200 25/25 [==============================] - 1s 27ms/step - loss: 0.0427 - accuracy: 0.9866 - val_loss: 0.2436 - val_accuracy: 0.9028 Epoch 81/200 25/25 [==============================] - 1s 23ms/step - loss: 0.0431 - accuracy: 0.9866 - val_loss: 0.2287 - val_accuracy: 0.9182 Epoch 82/200 25/25 [==============================] - 1s 24ms/step - loss: 0.0435 - accuracy: 0.9821 - val_loss: 0.2463 - val_accuracy: 0.9054 Epoch 83/200 25/25 [==============================] - 1s 25ms/step - loss: 0.0366 - accuracy: 0.9930 - val_loss: 0.2245 - val_accuracy: 0.9105 Epoch 84/200 25/25 [==============================] - 1s 27ms/step - loss: 0.0382 - accuracy: 0.9878 - val_loss: 0.2309 - val_accuracy: 0.9079 Epoch 85/200 25/25 [==============================] - 1s 28ms/step - loss: 0.0500 - accuracy: 0.9795 - val_loss: 0.2130 - val_accuracy: 0.9105 Epoch 86/200 25/25 [==============================] - 1s 26ms/step - loss: 0.0394 - accuracy: 0.9885 - val_loss: 0.3004 - val_accuracy: 0.8951 Epoch 87/200 25/25 [==============================] - 1s 25ms/step - loss: 0.0497 - accuracy: 0.9802 - val_loss: 0.2241 - val_accuracy: 0.9182 Epoch 88/200 25/25 [==============================] - 1s 21ms/step - loss: 0.0414 - accuracy: 0.9840 - val_loss: 0.2138 - val_accuracy: 0.9130 Epoch 89/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0412 - accuracy: 0.9878 - val_loss: 0.2477 - val_accuracy: 0.9028 Epoch 90/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0425 - accuracy: 0.9853 - val_loss: 0.2369 - val_accuracy: 0.9079 Epoch 91/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0408 - accuracy: 0.9853 - val_loss: 0.2769 - val_accuracy: 0.9054 Epoch 92/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0333 - accuracy: 0.9930 - val_loss: 0.2507 - val_accuracy: 0.9079 Epoch 93/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0404 - accuracy: 0.9853 - val_loss: 0.2095 - val_accuracy: 0.9130 Epoch 94/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0295 - accuracy: 0.9930 - val_loss: 0.2165 - val_accuracy: 0.9156 Epoch 95/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0285 - accuracy: 0.9942 - val_loss: 0.2600 - val_accuracy: 0.9028 Epoch 96/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0331 - accuracy: 0.9891 - val_loss: 0.2501 - val_accuracy: 0.9079 Epoch 97/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0310 - accuracy: 0.9891 - val_loss: 0.2275 - val_accuracy: 0.9105 Epoch 98/200 25/25 [==============================] - 1s 22ms/step - loss: 0.0358 - accuracy: 0.9898 - val_loss: 0.2817 - val_accuracy: 0.9003 Epoch 99/200 25/25 [==============================] - 1s 24ms/step - loss: 0.0364 - accuracy: 0.9904 - val_loss: 0.2161 - val_accuracy: 0.9105 Epoch 100/200 25/25 [==============================] - 1s 21ms/step - loss: 0.0272 - accuracy: 0.9936 - val_loss: 0.2180 - val_accuracy: 0.9105 Epoch 101/200 25/25 [==============================] - 1s 21ms/step - loss: 0.0398 - accuracy: 0.9853 - val_loss: 0.2126 - val_accuracy: 0.9079 Epoch 102/200 25/25 [==============================] - 1s 21ms/step - loss: 0.0343 - accuracy: 0.9891 - val_loss: 0.2539 - val_accuracy: 0.9028 Epoch 103/200 25/25 [==============================] - 1s 21ms/step - loss: 0.0416 - accuracy: 0.9866 - val_loss: 0.2091 - val_accuracy: 0.9105 Epoch 104/200 25/25 [==============================] - 1s 21ms/step - loss: 0.0329 - accuracy: 0.9878 - val_loss: 0.2261 - val_accuracy: 0.9105 Epoch 105/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0243 - accuracy: 0.9936 - val_loss: 0.2284 - val_accuracy: 0.9105 Epoch 106/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0343 - accuracy: 0.9872 - val_loss: 0.2457 - val_accuracy: 0.9054 Epoch 107/200 25/25 [==============================] - 1s 22ms/step - loss: 0.0252 - accuracy: 0.9942 - val_loss: 0.2123 - val_accuracy: 0.9182 Epoch 108/200 25/25 [==============================] - 1s 22ms/step - loss: 0.0277 - accuracy: 0.9917 - val_loss: 0.2392 - val_accuracy: 0.9105 Epoch 109/200 25/25 [==============================] - 1s 24ms/step - loss: 0.0276 - accuracy: 0.9917 - val_loss: 0.2317 - val_accuracy: 0.9130 Epoch 110/200 25/25 [==============================] - 1s 22ms/step - loss: 0.0248 - accuracy: 0.9942 - val_loss: 0.2242 - val_accuracy: 0.9105 Epoch 111/200 25/25 [==============================] - 1s 22ms/step - loss: 0.0265 - accuracy: 0.9930 - val_loss: 0.2266 - val_accuracy: 0.9156 Epoch 112/200 25/25 [==============================] - 1s 21ms/step - loss: 0.0283 - accuracy: 0.9923 - val_loss: 0.2771 - val_accuracy: 0.9028 Epoch 113/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0240 - accuracy: 0.9936 - val_loss: 0.2472 - val_accuracy: 0.9105 Epoch 114/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0290 - accuracy: 0.9910 - val_loss: 0.2542 - val_accuracy: 0.9105 Epoch 115/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0374 - accuracy: 0.9872 - val_loss: 0.2145 - val_accuracy: 0.9156 Epoch 116/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0283 - accuracy: 0.9923 - val_loss: 0.2443 - val_accuracy: 0.9105 Epoch 117/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0227 - accuracy: 0.9949 - val_loss: 0.2392 - val_accuracy: 0.9079 Epoch 118/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0244 - accuracy: 0.9949 - val_loss: 0.2559 - val_accuracy: 0.9105 Epoch 119/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0241 - accuracy: 0.9930 - val_loss: 0.2137 - val_accuracy: 0.9130 Epoch 120/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0212 - accuracy: 0.9968 - val_loss: 0.2639 - val_accuracy: 0.9028 Epoch 121/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0267 - accuracy: 0.9917 - val_loss: 0.2512 - val_accuracy: 0.9054 Epoch 122/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0222 - accuracy: 0.9949 - val_loss: 0.2137 - val_accuracy: 0.9130 Epoch 123/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0263 - accuracy: 0.9923 - val_loss: 0.1876 - val_accuracy: 0.9233 Epoch 124/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0174 - accuracy: 0.9968 - val_loss: 0.2321 - val_accuracy: 0.9130 Epoch 125/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0247 - accuracy: 0.9930 - val_loss: 0.2205 - val_accuracy: 0.9156 Epoch 126/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0170 - accuracy: 0.9974 - val_loss: 0.2054 - val_accuracy: 0.9156 Epoch 127/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0158 - accuracy: 0.9981 - val_loss: 0.2169 - val_accuracy: 0.9182 Epoch 128/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0245 - accuracy: 0.9923 - val_loss: 0.2415 - val_accuracy: 0.9105 Epoch 129/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0241 - accuracy: 0.9930 - val_loss: 0.2949 - val_accuracy: 0.9003 Epoch 130/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0187 - accuracy: 0.9962 - val_loss: 0.2061 - val_accuracy: 0.9182 Epoch 131/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0239 - accuracy: 0.9917 - val_loss: 0.2088 - val_accuracy: 0.9156 Epoch 132/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0248 - accuracy: 0.9942 - val_loss: 0.2229 - val_accuracy: 0.9105 Epoch 133/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0214 - accuracy: 0.9923 - val_loss: 0.2331 - val_accuracy: 0.9079 Epoch 134/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0204 - accuracy: 0.9962 - val_loss: 0.2422 - val_accuracy: 0.9079 Epoch 135/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0207 - accuracy: 0.9942 - val_loss: 0.2483 - val_accuracy: 0.9105 Epoch 136/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0200 - accuracy: 0.9955 - val_loss: 0.2466 - val_accuracy: 0.9079 Epoch 137/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0207 - accuracy: 0.9942 - val_loss: 0.2148 - val_accuracy: 0.9182 Epoch 138/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0214 - accuracy: 0.9942 - val_loss: 0.2268 - val_accuracy: 0.9156 Epoch 139/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0155 - accuracy: 0.9962 - val_loss: 0.2667 - val_accuracy: 0.9028 Epoch 140/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0150 - accuracy: 0.9974 - val_loss: 0.2193 - val_accuracy: 0.9156 Epoch 141/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0141 - accuracy: 0.9974 - val_loss: 0.2151 - val_accuracy: 0.9156 Epoch 142/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0140 - accuracy: 0.9974 - val_loss: 0.2269 - val_accuracy: 0.9156 Epoch 143/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0213 - accuracy: 0.9923 - val_loss: 0.2182 - val_accuracy: 0.9156 Epoch 144/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0123 - accuracy: 0.9968 - val_loss: 0.2269 - val_accuracy: 0.9182 Epoch 145/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0132 - accuracy: 0.9968 - val_loss: 0.1980 - val_accuracy: 0.9258 Epoch 146/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0186 - accuracy: 0.9955 - val_loss: 0.2100 - val_accuracy: 0.9156 Epoch 147/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0176 - accuracy: 0.9962 - val_loss: 0.2178 - val_accuracy: 0.9130 Epoch 148/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0201 - accuracy: 0.9955 - val_loss: 0.3188 - val_accuracy: 0.9003 Epoch 149/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0187 - accuracy: 0.9968 - val_loss: 0.2354 - val_accuracy: 0.9105 Epoch 150/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0178 - accuracy: 0.9968 - val_loss: 0.2513 - val_accuracy: 0.9105 Epoch 151/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0142 - accuracy: 0.9955 - val_loss: 0.2209 - val_accuracy: 0.9156 Epoch 152/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0139 - accuracy: 0.9968 - val_loss: 0.2021 - val_accuracy: 0.9233 Epoch 153/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0141 - accuracy: 0.9962 - val_loss: 0.2111 - val_accuracy: 0.9233 Epoch 154/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0134 - accuracy: 0.9962 - val_loss: 0.2025 - val_accuracy: 0.9207 Epoch 155/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0094 - accuracy: 0.9994 - val_loss: 0.1835 - val_accuracy: 0.9258 Epoch 156/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0125 - accuracy: 0.9968 - val_loss: 0.2072 - val_accuracy: 0.9233 Epoch 157/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0127 - accuracy: 0.9968 - val_loss: 0.2032 - val_accuracy: 0.9233 Epoch 158/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0149 - accuracy: 0.9968 - val_loss: 0.1958 - val_accuracy: 0.9233 Epoch 159/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0143 - accuracy: 0.9962 - val_loss: 0.2279 - val_accuracy: 0.9156 Epoch 160/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0120 - accuracy: 0.9981 - val_loss: 0.2207 - val_accuracy: 0.9182 Epoch 161/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0121 - accuracy: 0.9981 - val_loss: 0.2151 - val_accuracy: 0.9182 Epoch 162/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0117 - accuracy: 0.9968 - val_loss: 0.2462 - val_accuracy: 0.9105 Epoch 163/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0114 - accuracy: 0.9981 - val_loss: 0.1876 - val_accuracy: 0.9258 Epoch 164/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0108 - accuracy: 0.9981 - val_loss: 0.1967 - val_accuracy: 0.9258 Epoch 165/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0112 - accuracy: 0.9981 - val_loss: 0.2111 - val_accuracy: 0.9233 Epoch 166/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0102 - accuracy: 0.9994 - val_loss: 0.1902 - val_accuracy: 0.9284 Epoch 167/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0167 - accuracy: 0.9968 - val_loss: 0.1851 - val_accuracy: 0.9258 Epoch 168/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0128 - accuracy: 0.9962 - val_loss: 0.2213 - val_accuracy: 0.9233 Epoch 169/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0109 - accuracy: 0.9981 - val_loss: 0.2903 - val_accuracy: 0.9028 Epoch 170/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0111 - accuracy: 0.9974 - val_loss: 0.2402 - val_accuracy: 0.9105 Epoch 171/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0161 - accuracy: 0.9949 - val_loss: 0.2060 - val_accuracy: 0.9207 Epoch 172/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0114 - accuracy: 0.9974 - val_loss: 0.2200 - val_accuracy: 0.9182 Epoch 173/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0113 - accuracy: 0.9994 - val_loss: 0.2237 - val_accuracy: 0.9182 Epoch 174/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0127 - accuracy: 0.9968 - val_loss: 0.2500 - val_accuracy: 0.9079 Epoch 175/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0096 - accuracy: 0.9981 - val_loss: 0.2170 - val_accuracy: 0.9233 Epoch 176/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0097 - accuracy: 0.9987 - val_loss: 0.2057 - val_accuracy: 0.9284 Epoch 177/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0079 - accuracy: 0.9994 - val_loss: 0.2506 - val_accuracy: 0.9079 Epoch 178/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0098 - accuracy: 0.9974 - val_loss: 0.2317 - val_accuracy: 0.9207 Epoch 179/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0110 - accuracy: 0.9987 - val_loss: 0.2224 - val_accuracy: 0.9233 Epoch 180/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0100 - accuracy: 0.9994 - val_loss: 0.2070 - val_accuracy: 0.9284 Epoch 181/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0097 - accuracy: 0.9974 - val_loss: 0.2412 - val_accuracy: 0.9182 Epoch 182/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0115 - accuracy: 0.9942 - val_loss: 0.2618 - val_accuracy: 0.9079 Epoch 183/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0172 - accuracy: 0.9942 - val_loss: 0.2077 - val_accuracy: 0.9233 Epoch 184/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0106 - accuracy: 0.9962 - val_loss: 0.2246 - val_accuracy: 0.9258 Epoch 185/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0191 - accuracy: 0.9942 - val_loss: 0.1825 - val_accuracy: 0.9284 Epoch 186/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0091 - accuracy: 0.9994 - val_loss: 0.2279 - val_accuracy: 0.9233 Epoch 187/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0094 - accuracy: 0.9987 - val_loss: 0.2784 - val_accuracy: 0.9079 Epoch 188/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0114 - accuracy: 0.9974 - val_loss: 0.2485 - val_accuracy: 0.9156 Epoch 189/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0086 - accuracy: 0.9981 - val_loss: 0.2095 - val_accuracy: 0.9258 Epoch 190/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0083 - accuracy: 0.9994 - val_loss: 0.2127 - val_accuracy: 0.9233 Epoch 191/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0095 - accuracy: 0.9981 - val_loss: 0.2281 - val_accuracy: 0.9233 Epoch 192/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0080 - accuracy: 0.9994 - val_loss: 0.2699 - val_accuracy: 0.9054 Epoch 193/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0084 - accuracy: 0.9987 - val_loss: 0.2490 - val_accuracy: 0.9130 Epoch 194/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0062 - accuracy: 0.9987 - val_loss: 0.2378 - val_accuracy: 0.9182 Epoch 195/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0101 - accuracy: 0.9974 - val_loss: 0.2552 - val_accuracy: 0.9105 Epoch 196/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0082 - accuracy: 0.9981 - val_loss: 0.2523 - val_accuracy: 0.9156 Epoch 197/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0069 - accuracy: 0.9994 - val_loss: 0.2437 - val_accuracy: 0.9156 Epoch 198/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0102 - accuracy: 0.9974 - val_loss: 0.2234 - val_accuracy: 0.9258 Epoch 199/200 25/25 [==============================] - 0s 20ms/step - loss: 0.0066 - accuracy: 0.9994 - val_loss: 0.2790 - val_accuracy: 0.9079 Epoch 200/200 25/25 [==============================] - 1s 20ms/step - loss: 0.0072 - accuracy: 0.9987 - val_loss: 0.2198 - val_accuracy: 0.9258
After 200 training epochs, we will plot the accuracy and loss curves:
In [ ]:
plt.plot(history.history['accuracy'])
plt.plot(history.history['val_accuracy'])
plt.title('model accuracy')
plt.ylabel('accuracy')
plt.xlabel('epoch')
plt.legend(['train', 'val'], loc='lower right')
plt.show()
In [ ]:
plt.plot(history.history['loss'])
plt.plot(history.history['val_loss'])
plt.title('model loss')
plt.ylabel('loss')
plt.xlabel('epoch')
plt.legend(['train', 'val'], loc='upper right')
plt.show()
We can save both the architecture and the trained weights for later use.
In [ ]:
model_json = Model.to_json()
with open("drive/My Drive/Datasets/Car_detection/Model200_new.json", "w") as json_file:
json_file.write(model_json)
# serialize weights to HDF5
Model.save_weights("drive/My Drive/Datasets/Car_detection/Model200_weights_new.weights.h5")
In [ ]:
from keras.models import model_from_json
In [ ]:
json_file = open('drive/My Drive/Datasets/Car_detection/Model200_new.json', 'r')
loaded_model_json = json_file.read()
json_file.close()
loaded_model = model_from_json(loaded_model_json)
# load weights into new model
loaded_model.load_weights("drive/My Drive/Datasets/Car_detection/Model200_weights_new.weights.h5")
Now let's apply our model to an entire image. To make this possible, we will create functions to cycle through and divide the image into sizes equal to the one we trained and also resize the image.
In [ ]:
def resize(image, width=None, height=None, inter=cv2.INTER_AREA):
dim = None
(h, w) = image.shape[:2]
if width is None and height is None:
return image
if width is None:
r = height / float(h)
dim = (int(w * r), height)
else:
r = width / float(w)
dim = (width, int(h * r))
resized = cv2.resize(image, dim, interpolation=inter)
return resized
In [ ]:
def pyramid(image, scale=1.5, minSize=(30, 30)):
yield image
while True:
w = int(image.shape[1] / scale)
image = resize(image, width=w)
if image.shape[0] < minSize[1] or image.shape[1] < minSize[0]:
break
yield image
def sliding_window(image, stepSize, windowSize):
for y in range(0, image.shape[0], stepSize):
for x in range(0, image.shape[1], stepSize):
yield (x, y, image[y:y + windowSize[1], x:x + windowSize[0]])
We can import our test image:
In [ ]:
test_path = 'drive/My Drive/Datasets/Car_detection/Example_cars_detect.tif'
In [ ]:
src_test = rasterio.open(test_path)
im_test = src_test.read()
In [ ]:
im_test = im_test.transpose([1,2,0])
In [ ]:
im_test = im_test.astype('uint8')
In [ ]:
image = im_test[:,:,0:3]
Let's set the parameters and apply the sliding window to our image. If the model detects cars in the window, the boundary of that window will be added to the list of bboxes:
In [ ]:
scale = 1
minSize = (4000, 4000)
# Sliding window parameters
stepSize = 20
(winW, winH) = (80, 80)
In [ ]:
bboxes = np.zeros(4,np.int64)
for i, resized in enumerate(pyramid(image, scale=scale, minSize=minSize)):
for (x, y, window) in sliding_window(resized, stepSize=stepSize, windowSize=(winW, winH)):
if (window.shape[0] == winH and window.shape[1] == winW):
clone = resized.copy()
cropped_img = resized[y:y + winH, x:x + winW]
cropped_img = cropped_img[np.newaxis,:,:,:]
cropped_img = cropped_img/255
predict = loaded_model.predict(cropped_img)
predict = np.round(predict).astype(np.uint8)
y_pred = predict[0][0]
if y_pred == 1:
print('detected')
if i != 0:
bboxes = np.vstack((bboxes, np.array([
int(x*scale*i), int(y*scale*i),
int((x + winW)*scale*i), int((y + winH)*scale*i)])))
else:
bboxes = np.vstack((bboxes, np.array([
int(x),int(y),int(x + winW), int(y + winH)])))
detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected detected
Let's plot the results:
In [ ]:
bboxes = np.delete(bboxes, (0), axis=0)
In [ ]:
plt.figure(figsize=[20,20])
plt.imshow(im_test)
for box in bboxes:
rect = plt.Rectangle((box[0], box[1]), box[2]-box[0], box[3]-box[1], edgecolor='r', facecolor='none')
plt.gca().add_patch(rect)
We saw that we have regions with several bboxes close to each other. Let's use non-maximum suppression to remove inaccurate detections from the same region:
In [ ]:
def non_max_suppression_fast(boxes, overlapThresh):
'''
This function performs non maximal supression over a set of bounding boxes
where the resulting boxes contain most of the information of the initial
boxes.
Args:
boxes(numpy.array): Set of bounding boxes of shape=(None,4).
overlapThresh(float): Supression threshold for the overlapped area
between boxes.
Returns:
boxes(numpy.array): Resulting boxes after performing the NMS of
shape=(None,4).
'''
if len(boxes) == 0:
return []
if boxes.dtype.kind == "i":
boxes = boxes.astype("float")
pick = []
x1 = boxes[:,0]
y1 = boxes[:,1]
x2 = boxes[:,2]
y2 = boxes[:,3]
area = (x2 - x1 + 1) * (y2 - y1 + 1)
idxs = np.argsort(y2)
while len(idxs) > 0:
last = len(idxs) - 1
i = idxs[last]
pick.append(i)
xx1 = np.maximum(x1[i], x1[idxs[:last]])
yy1 = np.maximum(y1[i], y1[idxs[:last]])
xx2 = np.minimum(x2[i], x2[idxs[:last]])
yy2 = np.minimum(y2[i], y2[idxs[:last]])
# compute the width and height of the bounding box
w = np.maximum(0, xx2 - xx1 + 1)
h = np.maximum(0, yy2 - yy1 + 1)
overlap = (w * h) / area[idxs[:last]]
idxs = np.delete(idxs, np.concatenate(([last],
np.where(overlap > overlapThresh)[0])))
return boxes[pick].astype("int")
In [ ]:
nms_bboxes = non_max_suppression_fast(bboxes, 0.3)
In [ ]:
plt.figure(figsize=[20,20])
plt.imshow(im_test)
for box in nms_bboxes:
rect = plt.Rectangle((box[0], box[1]), box[2]-box[0], box[3]-box[1], edgecolor='r', facecolor='none')
plt.gca().add_patch(rect)
For each detection found, we will obtain the centroid and convert it to latitude and longitude values:
In [ ]:
ls_x = []
ls_y = []
for i in nms_bboxes:
xcenter = (i[2] - i[0])/2 + i[0]
ycenter = (i[3] - i[1])/2 + i[1]
xcenter = np.round(xcenter)
ycenter = np.round(ycenter)
xs, ys = rasterio.transform.xy(src_test.transform, ycenter, xcenter)
ls_x.append(xs)
ls_y.append(ys)
In [ ]:
df_xy = pd.DataFrame([])
df_xy['x'] = ls_x
df_xy['y'] = ls_y
Finally, we create a geodataframe of points:
In [ ]:
gdf = gpd.GeoDataFrame(df_xy, geometry=gpd.points_from_xy(df_xy['x'], df_xy['y']))
In [ ]:
gdf = gdf.set_crs(src_test.crs.to_dict()['init'])
In [ ]:
fig, ax = plt.subplots(figsize=(20, 20))
show((src_test,[1,2,3]), ax=ax)
gdf.plot(ax=ax, marker='o', color='red', markersize=15)
Out[ ]:
<matplotlib.axes._subplots.AxesSubplot at 0x7f89d4e259d0>
We save it in a .json file and download:
In [ ]:
gdf.to_file('/content/cars_locations.json', driver="GeoJSON")
In [ ]:
from google.colab import files
files.download('/content/cars_locations.json')