Added missing Google Static Maps API URL

.gitignore

@@ -0,0 +1,8 @@
+/download_area/results
+output/
+test/
+.project
+.pydevproject
+Dataset.h5
+*.pyc
+Classifications.p

common.py

@@ -0,0 +1,24 @@
+from keras.models import Sequential
+from keras.layers import Conv2D, MaxPooling2D
+from keras.layers import Activation, Dropout, Flatten, Dense
+
+
+def getModel():
+    keras_model = Sequential()
+    keras_model.add(Conv2D(64, (3, 3), input_shape=(256,256,3)))
+    keras_model.add(Activation('relu'))
+    keras_model.add(MaxPooling2D(pool_size=(2, 2)))
+    keras_model.add(Conv2D(64, (3, 3)))
+    keras_model.add(Activation('relu'))
+    keras_model.add(MaxPooling2D(pool_size=(2, 2)))
+    keras_model.add(Conv2D(64, (3, 3)))
+    keras_model.add(Activation('relu'))
+    keras_model.add(MaxPooling2D(pool_size=(2, 2)))
+    keras_model.add(Flatten())
+    keras_model.add(Dense(64,activation='relu'))
+    keras_model.add(Dropout(0.5))
+    keras_model.add(Dense(3, activation='softmax')) #3 classes
+    return keras_model
+
+dataset_name = 'Dataset.h5'
+classifications_file_name = 'Classifications.p'

create_hdf5.py

@@ -2,6 +2,11 @@
 import pickle
 import os
 import argparse
+import common
+from PIL import Image
+import numpy as np
+import imutils
+
 
 parser = argparse.ArgumentParser()
 parser.add_argument('ImagesDirectory',help='Path to images directory')
@@ -20,10 +25,9 @@ def check_args():
 		print "Loading."
 
 check_args()
-dataset_h5_path ='Dataset.h5'
 
-if not os.path.exists(dataset_h5_path):
-    dataset_h5 = h5py.File(dataset_h5_path, 'a')
+if not os.path.exists(common.dataset_name):
+    dataset_h5 = h5py.File(common.dataset_name, 'a')
     dataset_h5.create_dataset('data', shape=(1731, 256,256,3),
                               maxshape=(None, 256,256,3),
                               compression='gzip', compression_opts=6)
@@ -33,10 +37,10 @@ def check_args():
     print ('Dataset was created')
 else:
     print ('Dataset exists')
-    dataset_h5 = h5py.File(dataset_h5_path, 'a')
+    dataset_h5 = h5py.File(common.dataset_name, 'a')
 
 images_directory_path = args.ImagesDirectory
-pickled_data = 'Classifications.p'
+pickled_data = common.classifications_file_name
 
 
 with open(pickled_data, 'rb') as f:
@@ -49,9 +53,6 @@ def check_args():
 all_labels = enc.transform(cleaned_source.values())
 
 
-from PIL import Image
-import numpy as np
-import imutils
 
 
 for i in range(0, len(cleaned_source.items())//16, 16):

download_area/download_area.rb

@@ -124,7 +124,8 @@
 end
 #--- 
 def build_url(lat,lng,zoom=DEFAULT_ZOOM, size=IMAGE_SIZE)
-  raise("No URL Defined")
+  raise("Please uncomment and enter your Google API KEY below")
+  # "https://maps.googleapis.com/maps/api/staticmap?center=#{lat},#{lng}&zoom=#{zoom}&maptype=satellite&size=#{size}x#{size}&key=YOUR_GOOGLE_API_KEY" 
 end
 
 lookup = {}

gui.py

@@ -1,5 +1,5 @@
 from Tkinter import *
-from PIL import Image, ImageTk, ImageDraw
+from PIL import ImageTk, ImageDraw
 import glob
 
 global index, classifications

model_predict.py

@@ -1,7 +1,7 @@
 import os, glob, pickle, argparse, imutils
 import numpy as np
 from PIL import Image
-
+import common
 
 weights = glob.glob('output/weights*.h5')
 parser = argparse.ArgumentParser()
@@ -28,8 +28,8 @@ def process_img(image):
 
 check_args()
 
-import model_define
-classifier = model_define.create_model_v3()
+
+classifier = common.getModel()
 classifier.load_weights(weights[0])
 
 images_directory = args.ImagesDirectory

model_train.py

@@ -5,57 +5,44 @@
 import pickle
 import h5py
 import numpy as np
-import model_define
+import common
+from collections import OrderedDict
+import matplotlib.pyplot as plt
 
-file_path = 'Dataset.h5'
-dataset_h5 = h5py.File(file_path, 'r')
+# hyper-parameters
+epochs = 500
+batch_size = 16
+learning_rate_init = 0.00001
+loss_function = 'categorical_crossentropy'
+optimizer = 'adam'
 
-print ('H5 Dataset retrieved')
 
+dataset_h5 = h5py.File(common.dataset_name, 'r')
 labels = {0:'coconut farm', 1:'partly coconut farm', 2:'not a coconut farm'}
-
-with open('Classifications.p', 'rb') as f:
+with open(common.classifications_file_name, 'rb') as f:
     datasource = pickle.load(f)
-
-print ('Paths and labels unpickled.')
-from collections import OrderedDict
 cleaned_source = OrderedDict((k,v-1) for k,v in datasource.items() if 1<=v<=3)
-
 X = cleaned_source.keys()
 dataset_labels = [v for v in cleaned_source.values()]
-# dimensions of our images.
-img_width, img_height = 256, 256
-
-epochs = 500
-batch_size = 16
 
 data_indices =[i for i in range(len(cleaned_source.items()))]
-
 train, test, _, _= train_test_split(data_indices, dataset_labels, train_size=0.75, random_state=42)
 train.sort()
 test.sort()
-
 x = np.asarray(train)
 dataset_labels = np.asarray(dataset_labels)
 temp= dataset_labels[x]
-classes = np.array([0,1,2])
+classes = np.array(labels.keys())
 class_weight_vect = compute_class_weight('balanced', classes, temp)
 class_weight = {0:class_weight_vect[0], 1:class_weight_vect[1], 2:class_weight_vect[2]}
-
 indices = {'train':train, 'test':test}
-
-model_name = '15_July_Model_v3'
-
+model_name = 'Coconut-Not-Coconut'
 with open('output/'+model_name+'_indices.p', 'wb') as f:
     pickle.dump(indices, f, protocol=2)
+model = common.getModel()
+model.compile(loss=loss_function,optimizer=optimizer, metrics=['accuracy'])
 
-print ("Building model - ", model_name)
-# create model
-model = model_define.create_model_v3()
-model.compile(loss='categorical_crossentropy',optimizer='adam', metrics=['accuracy'])
-print ("Model built.")
-
-#create augmented data for training
+#note: model doesn't converge unless we re-scale pixels to range [0,1]
 train_datagen = ImageDataGenerator(fill_mode='reflect',
                                    shear_range=0.3,
                                    rotation_range=90,
@@ -65,27 +52,34 @@
 
 train_generator = train_datagen.flow(dataset_h5['data'][train, :,:,:], dataset_h5['labels'][train,:],
 batch_size=batch_size)
-
-#create validation data
 test_datagen = ImageDataGenerator(rescale=1. / 255)
-
 validation_generator = test_datagen.flow(dataset_h5['data'][test,:,:,:], dataset_h5['labels'][test,:],
 batch_size=batch_size)
-
-print ("Training and Evaluating model")
-
 filepath = "output/weights"+model_name+".h5"
-
 checkpoint = callbacks.ModelCheckpoint(filepath, monitor='val_acc', verbose=1, save_best_only=True, mode='max')
-#early_stop = callbacks.EarlyStopping(monitor='val_loss', min_delta=0.00001, patience=4, mode='min', verbose=1)
-reduce_lr = callbacks.ReduceLROnPlateau(monitor='val_loss', factor=0.8,patience=5, min_lr=0.00001, verbose=1)
-
+reduce_lr = callbacks.ReduceLROnPlateau(monitor='val_loss', factor=0.8,patience=5, min_lr=learning_rate_init, verbose=1)
 callbacks_list = [checkpoint, reduce_lr]
-
-model.fit_generator(train_generator,
+history = model.fit_generator(train_generator,
     steps_per_epoch=len(train) // batch_size,
     epochs=epochs,
     validation_data=validation_generator,
     validation_steps=len(test) // batch_size, callbacks=callbacks_list, verbose=1)
 
-print ('Model: '+model_name +' saved.')
+
+# summarize history for accuracy
+plt.plot(history.history['acc'])
+plt.plot(history.history['val_acc'])
+plt.title('model accuracy')
+plt.ylabel('accuracy')
+plt.xlabel('epoch')
+plt.legend(['train', 'test'], loc='upper left')
+plt.show()
+
+# summarize history for loss
+plt.plot(history.history['loss'])
+plt.plot(history.history['val_loss'])
+plt.title('model loss')
+plt.ylabel('loss')
+plt.xlabel('epoch')
+plt.legend(['train', 'test'], loc='upper left')
+plt.show()