Ben041 commited on
Commit
67dce95
·
verified ·
1 Parent(s): 7245d3f
Files changed (2) hide show
  1. soil_type_model.h5 +3 -0
  2. train_soil_classifier.py +205 -0
soil_type_model.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d28500d034cb138271a60a1ee91e31fc124ec7204eadc31a317f5f2e621a6e5f
3
+ size 116262336
train_soil_classifier.py ADDED
@@ -0,0 +1,205 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import os
2
+ import numpy as np
3
+ import tensorflow as tf
4
+ from tensorflow.keras import layers, models
5
+ from tensorflow.keras.preprocessing.image import ImageDataGenerator
6
+ from sklearn.model_selection import train_test_split
7
+ import matplotlib.pyplot as plt
8
+
9
+ # Define paths
10
+ dataset_path = r"C:\Users\Joel\Desktop\bin\Soil-Type-1\Soil Types"
11
+ output_model_path = r"C:\Users\Joel\Desktop\bin\Soil-Type-1\soil_type_model.h5"
12
+
13
+ def create_soil_type_classifier():
14
+ # 1. Organize and prepare the dataset
15
+
16
+ # Create train/validation directories if they don't exist
17
+ train_dir = os.path.join(dataset_path, 'train')
18
+ val_dir = os.path.join(dataset_path, 'validation')
19
+
20
+ if not os.path.exists(train_dir):
21
+ os.makedirs(train_dir)
22
+ if not os.path.exists(val_dir):
23
+ os.makedirs(val_dir)
24
+
25
+ # Create subdirectories for each soil type in train and validation folders
26
+ soil_types = [d for d in os.listdir(dataset_path) if os.path.isdir(os.path.join(dataset_path, d))
27
+ and d not in ['train', 'validation']]
28
+
29
+ for soil_type in soil_types:
30
+ train_soil_dir = os.path.join(train_dir, soil_type)
31
+ val_soil_dir = os.path.join(val_dir, soil_type)
32
+
33
+ if not os.path.exists(train_soil_dir):
34
+ os.makedirs(train_soil_dir)
35
+ if not os.path.exists(val_soil_dir):
36
+ os.makedirs(val_soil_dir)
37
+
38
+ # 2. Split data into training and validation sets (80/20 split)
39
+ for soil_type in soil_types:
40
+ soil_dir = os.path.join(dataset_path, soil_type)
41
+ image_files = [f for f in os.listdir(soil_dir) if f.endswith(('.jpg', '.jpeg', '.png'))]
42
+
43
+ # Shuffle image files
44
+ np.random.shuffle(image_files)
45
+
46
+ # Split the dataset (80% train, 20% validation)
47
+ train_files = image_files[:int(0.8 * len(image_files))]
48
+ val_files = image_files[int(0.8 * len(image_files)):]
49
+
50
+ # Copy files to train directory
51
+ import shutil
52
+ for file in train_files:
53
+ src = os.path.join(soil_dir, file)
54
+ dst = os.path.join(train_dir, soil_type, file)
55
+ shutil.copy(src, dst)
56
+
57
+ # Copy files to validation directory
58
+ for file in val_files:
59
+ src = os.path.join(soil_dir, file)
60
+ dst = os.path.join(val_dir, soil_type, file)
61
+ shutil.copy(src, dst)
62
+
63
+ print(f"Dataset organized into {len(soil_types)} categories: {', '.join(soil_types)}")
64
+
65
+ # 3. Set up data augmentation and preprocessing
66
+ train_datagen = ImageDataGenerator(
67
+ rescale=1./255,
68
+ rotation_range=20,
69
+ width_shift_range=0.2,
70
+ height_shift_range=0.2,
71
+ shear_range=0.2,
72
+ zoom_range=0.2,
73
+ horizontal_flip=True,
74
+ fill_mode='nearest'
75
+ )
76
+
77
+ validation_datagen = ImageDataGenerator(rescale=1./255)
78
+
79
+ # Set target image size
80
+ img_height, img_width = 224, 224
81
+ batch_size = 32
82
+
83
+ # Create data generators
84
+ train_generator = train_datagen.flow_from_directory(
85
+ train_dir,
86
+ target_size=(img_height, img_width),
87
+ batch_size=batch_size,
88
+ class_mode='categorical'
89
+ )
90
+
91
+ validation_generator = validation_datagen.flow_from_directory(
92
+ val_dir,
93
+ target_size=(img_height, img_width),
94
+ batch_size=batch_size,
95
+ class_mode='categorical'
96
+ )
97
+
98
+ # 4. Create and compile the model
99
+ model = models.Sequential([
100
+ # Base convolutional layers
101
+ layers.Conv2D(32, (3, 3), activation='relu', input_shape=(img_height, img_width, 3)),
102
+ layers.MaxPooling2D((2, 2)),
103
+ layers.Conv2D(64, (3, 3), activation='relu'),
104
+ layers.MaxPooling2D((2, 2)),
105
+ layers.Conv2D(128, (3, 3), activation='relu'),
106
+ layers.MaxPooling2D((2, 2)),
107
+ layers.Conv2D(128, (3, 3), activation='relu'),
108
+ layers.MaxPooling2D((2, 2)),
109
+
110
+ # Flatten and dense layers
111
+ layers.Flatten(),
112
+ layers.Dropout(0.5), # Reduce overfitting
113
+ layers.Dense(512, activation='relu'),
114
+ layers.Dense(len(soil_types), activation='softmax') # Output neurons equal to number of soil types
115
+ ])
116
+
117
+ # Compile the model
118
+ model.compile(
119
+ optimizer='adam',
120
+ loss='categorical_crossentropy',
121
+ metrics=['accuracy']
122
+ )
123
+
124
+ # 5. Train the model
125
+ epochs = 15
126
+
127
+ history = model.fit(
128
+ train_generator,
129
+ steps_per_epoch=train_generator.samples // batch_size,
130
+ epochs=epochs,
131
+ validation_data=validation_generator,
132
+ validation_steps=validation_generator.samples // batch_size
133
+ )
134
+
135
+ # 6. Save the model
136
+ model.save(output_model_path)
137
+ print(f"Model saved to {output_model_path}")
138
+
139
+ # 7. Plot training results
140
+ acc = history.history['accuracy']
141
+ val_acc = history.history['val_accuracy']
142
+ loss = history.history['loss']
143
+ val_loss = history.history['val_loss']
144
+
145
+ plt.figure(figsize=(12, 4))
146
+ plt.subplot(1, 2, 1)
147
+ plt.plot(acc, label='Training Accuracy')
148
+ plt.plot(val_acc, label='Validation Accuracy')
149
+ plt.xlabel('Epoch')
150
+ plt.ylabel('Accuracy')
151
+ plt.legend()
152
+
153
+ plt.subplot(1, 2, 2)
154
+ plt.plot(loss, label='Training Loss')
155
+ plt.plot(val_loss, label='Validation Loss')
156
+ plt.xlabel('Epoch')
157
+ plt.ylabel('Loss')
158
+ plt.legend()
159
+
160
+ plt.tight_layout()
161
+ plt.savefig(os.path.join(dataset_path, 'training_results.png'))
162
+ plt.show()
163
+
164
+ # Return model and class indices for later use
165
+ return model, train_generator.class_indices
166
+
167
+ def predict_soil_type(model, image_path, class_indices):
168
+ """
169
+ Predict soil type for a given image
170
+ """
171
+ img_height, img_width = 224, 224
172
+
173
+ # Load and preprocess the image
174
+ img = tf.keras.preprocessing.image.load_img(
175
+ image_path, target_size=(img_height, img_width)
176
+ )
177
+ img_array = tf.keras.preprocessing.image.img_to_array(img)
178
+ img_array = tf.expand_dims(img_array, 0) / 255.0
179
+
180
+ # Make prediction
181
+ predictions = model.predict(img_array)
182
+ score = tf.nn.softmax(predictions[0])
183
+
184
+ # Get the predicted class and confidence
185
+ predicted_class = np.argmax(score)
186
+ confidence = 100 * np.max(score)
187
+
188
+ # Map class index to soil type name (reverse the class_indices dictionary)
189
+ class_names = {v: k for k, v in class_indices.items()}
190
+ predicted_soil_type = class_names[predicted_class]
191
+
192
+ print(f"This image is classified as: {predicted_soil_type} with {confidence:.2f}% confidence")
193
+ return predicted_soil_type, confidence
194
+
195
+ # Run the model creation and training
196
+ if __name__ == "__main__":
197
+ model, class_indices = create_soil_type_classifier()
198
+
199
+ # Example of how to use the model for prediction
200
+ # (you would replace this with your own test image)
201
+ test_image = input("Enter path to a test image: ")
202
+ if os.path.exists(test_image):
203
+ predict_soil_type(model, test_image, class_indices)
204
+ else:
205
+ print("Invalid image path. You can test the model later with a valid image.")