NOTE: This Use Case is not purposed for resource constrained devices.
Crop weeds are weeds that grow amongst crops. Despite the potential for some crop weeds to be used as a food source, many can also prove harmful to crops, both directly and indirectly. Crop weeds can inhibit the growth of crops, contaminate harvested crops and often spread rapidly.And hence it is necessary for the farmers to detect weed and remove them in order to protect their plants and crops.
Importing the dataset¶
In [1]:
!wget -N "https://cainvas-static.s3.amazonaws.com/media/user_data/cainvas-admin/Weed.zip"
!unzip -qo Weed.zip
!rm Weed.zip
Importing the dependencies¶
In [2]:
import numpy as np
import pandas as pd
import os
from os import listdir
import tensorflow as tf
from keras.preprocessing.image import ImageDataGenerator
import cv2
import matplotlib.pyplot as plt
%matplotlib inline
import imutils
from keras.layers import Dense, Flatten, AveragePooling2D, Dropout
from keras.models import Model
from keras.applications.vgg16 import VGG16
from keras.preprocessing import image
from keras.preprocessing.image import ImageDataGenerator
from keras.optimizers import Adam
from tensorflow.keras.models import Model,load_model
from sklearn.utils import shuffle
from tensorflow.keras.models import Sequential
from keras.layers import Reshape
from sklearn.model_selection import train_test_split
In [3]:
# Defining the dataset directory
image_dir="weed-no-weed/"
Creating a new folder to store augmented images¶
In [4]:
# Defining t=directory to store augmented images
os.makedirs('weeds-not-weeds/augmented-images')
os.makedirs('weeds-not-weeds/augmented-images/yes')
os.makedirs('weeds-not-weeds/augmented-images/no')
In [5]:
def augment_data(file_dir, n_generated_samples, save_to_dir):
data_gen = ImageDataGenerator(rotation_range=10,
width_shift_range=0.1,
height_shift_range=0.1,
shear_range=0.1,
brightness_range=(0.3, 1.0),
horizontal_flip=True,
vertical_flip=True,
fill_mode='nearest'
)
for filename in listdir(file_dir):
image = cv2.imread(file_dir + '/' + filename)
# reshape the image
image = image.reshape((1,)+image.shape)
save_prefix = 'aug_' + filename[:-4]
i=0
for batch in data_gen.flow(x=image, batch_size=1, save_to_dir=save_to_dir,save_prefix=save_prefix, save_format='jpg'):
i += 1
if i > n_generated_samples:
break
In [6]:
augmented_data_path ='weeds-not-weeds/augmented-images/'
augment_data(file_dir=image_dir+'weed',n_generated_samples=6, save_to_dir=augmented_data_path+'yes')
augment_data(file_dir=image_dir+'not_weed', n_generated_samples=9, save_to_dir=augmented_data_path+'no')
Loading the data and preparing it for Model Training¶
In [7]:
def load_data(dir_list, image_size):
# load all images in a directory
X = []
y = []
image_width, image_height = image_size
for directory in dir_list:
for filename in listdir(directory):
image = cv2.imread(directory+'/'+filename)
#image = crop_brain_contour(image, plot=False)
image = cv2.resize(image, dsize=(image_width, image_height), interpolation=cv2.INTER_CUBIC)
# normalize values
image = image / 255.
# convert image to numpy array and append it to X
X.append(image)
# append a value of 1 to the target array if the image
# is in the folder named 'yes', otherwise append 0.
if directory[-3:] == 'yes':
y.append([1])
else:
y.append([0])
X = np.array(X)
y = np.array(y)
# Shuffle the data
X, y = shuffle(X, y)
print(f'Number of examples is: {len(X)}')
print(f'X shape is: {X.shape}')
print(f'y shape is: {y.shape}')
return X, y
In [8]:
augmented_yes =augmented_data_path+'yes'
augmented_no = augmented_data_path+'no'
IMG_WIDTH, IMG_HEIGHT = (224, 224)
X, y = load_data([augmented_yes, augmented_no], (IMG_WIDTH, IMG_HEIGHT))
Data Visualization¶
In [9]:
def plot_sample_images(X, y, n=40):
for label in [0,1]:
# grab the first n images with the corresponding y values equal to label
images = X[np.argwhere(y == label)]
n_images = images[:n]
columns_n = 10
rows_n = int(n/ columns_n)
plt.figure(figsize=(10, 8))
i = 1 # current plot
for image in n_images:
plt.subplot(rows_n, columns_n, i)
plt.imshow(image[0])
# remove ticks
plt.tick_params(axis='both', which='both',
top=False, bottom=False, left=False, right=False,
labelbottom=False, labeltop=False, labelleft=False, labelright=False)
i += 1
label_to_str = lambda label: "Yes" if label == 1 else "No"
plt.suptitle(f"Weed Crop: {label_to_str(label)}")
plt.show()
In [10]:
plot_sample_images(X, y)
Test-Train Split¶
In [11]:
def split_data(X, y, test_size=0.2):
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=test_size)
return X_train, y_train, X_test, y_test
In [12]:
X_train, y_train, X_test, y_test = split_data(X, y, test_size=0.15)
In [13]:
print ("number of training examples = " + str(X_train.shape[0]))
print ("number of test examples = " + str(X_test.shape[0]))
Defining Model Architecture¶
In [14]:
# Model creation with changes
model = VGG16(input_shape=(224,224,3),include_top=False)
for layer in model.layers:
layer.trainable = False
newModel = model.output
newModel = AveragePooling2D()(newModel)
newModel = Flatten()(newModel)
newModel = Dense(128, activation="relu")(newModel)
newModel = Dropout(0.5)(newModel)
newModel = Dense(1, activation='sigmoid')(newModel)
model = Model(inputs=model.input, outputs=newModel)
In [15]:
model.summary()
In [16]:
opt=Adam(learning_rate=0.0001)
model.compile(optimizer=opt, loss='binary_crossentropy', metrics=['accuracy'])
Model Training¶
In [17]:
history = model.fit(x=X_train, y=y_train, batch_size=32, epochs=10, validation_data=(X_test, y_test))
Training Plots¶
In [18]:
# Loss Curves
plt.figure(figsize=[14,10])
plt.subplot(211)
plt.plot(history.history['loss'],'r',linewidth=3.0)
plt.plot(history.history['val_loss'],'b',linewidth=3.0)
plt.legend(['Training loss', 'Validation Loss'],fontsize=18)
plt.xlabel('Epochs ',fontsize=16)
plt.ylabel('Loss',fontsize=16)
plt.title('Loss Curves',fontsize=16)
# Accuracy Curves
plt.figure(figsize=[14,10])
plt.subplot(212)
plt.plot(history.history['accuracy'],'r',linewidth=3.0)
plt.plot(history.history['val_accuracy'],'b',linewidth=3.0)
plt.legend(['Training Accuracy', 'Validation Accuracy'],fontsize=18)
plt.xlabel('Epochs ',fontsize=16)
plt.ylabel('Accuracy',fontsize=16)
plt.title('Accuracy Curves',fontsize=16)
Out[18]:
In [19]:
model.save("Weed.h5")
Accessing the performance of the Model¶
In [20]:
image = cv2.imread("weed-no-weed/not_weed/11858d2a42ad30b668f0764530a00a67.jpg")
image = cv2.resize(image, dsize=(IMG_WIDTH, IMG_HEIGHT), interpolation=cv2.INTER_CUBIC)
# normalize values
image = image / 255.
image = np.array(image)
test_image = np.expand_dims(image,axis=0)
In [21]:
result = model.predict(test_image)
if result > 0.5:
print("Weed")
else:
print("Not Weed")
plt.imshow(image)
Out[21]:
In [22]:
image = cv2.imread("weed-no-weed/weed/6aa7fa3ff04411b410c46c9213575dc6.jpg")
image = cv2.resize(image, dsize=(IMG_WIDTH, IMG_HEIGHT), interpolation=cv2.INTER_CUBIC)
# normalize values
image = image / 255.
image = np.array(image)
test_image = np.expand_dims(image,axis=0)
In [23]:
result = model.predict(test_image)
if result > 0.5:
print("Weed")
else:
print("Not Weed")
plt.imshow(image)
Out[23]:
Compiling the Model with DeepC Compiler¶
In [24]:
!deepCC Weed.h5