EarthFinesse: Military Terrain Classifier

EarthFinesse is a high-accuracy military terrain classifier powered by deep learning. It classifies terrain types such as Grassy, Marshy, Rocky, and Sandy with an accuracy of over 97.87%, setting a new benchmark in this domain. The model uses the MobileNetV2 architecture, optimized for efficient and accurate terrain classification.

Table of Contents

• Installation
• Usage
• Model Training
• Training Procedure
• Training Results
• Applications
• Contributing
• License

Installation

1. Clone the repository:

   git clone https://github.com/PiPlusTheta/EarthFinesse.git
   cd EarthFinesse

2. Install the required Python packages:

   pip install -r requirements.txt

Usage

Streamlit User Interface

EarthFinesse comes with a user-friendly Streamlit interface for bulk image classification. Run the following command to start the application:

streamlit run app.py

Upload reconnaissance images, and the model will classify them into terrain types with confidence scores.

Model Inference

To perform individual image classification using the trained model, use the following code snippet:

# Load the model
from tensorflow.keras.models import load_model

model = load_model('terrain__2023_09_13__11_52_06___Accuracy_0.9787.h5')

# Load and preprocess the image
from tensorflow.keras.preprocessing import image
from tensorflow.keras.applications.mobilenet_v2 import preprocess_input
import numpy as np

img_path = 'path_to_image.jpg'
img = image.load_img(img_path, target_size=(224, 224))
img = image.img_to_array(img)
img = preprocess_input(img)
img = np.expand_dims(img, axis=0)

# Perform inference
prediction = model.predict(img)
label_index = np.argmax(prediction)
terrain_label = {0: 'Grassy', 1: 'Marshy', 2: 'Rocky', 3: 'Sandy'}[label_index]
confidence = prediction[0, label_index]

print(f"Predicted Terrain: {terrain_label}")
print(f"Confidence: {confidence * 100:.2f}%")

Model Training

Dataset

The model was trained on a dataset consisting of 45.1k images, with more than 10k images for each terrain class (Grassy, Marshy, Rocky, Sandy).

Training Procedure

Data Augmentation

The training data is augmented using techniques like shear, zoom, and horizontal flip to increase diversity.

MobileNetV2 Base Model

The MobileNetV2 architecture, pre-trained on ImageNet, is used as the base model for feature extraction. All base model layers are frozen to retain pre-trained knowledge.

Custom Classification Head

A custom classification head is added to the base model. It includes a global average pooling layer, a dense layer with 1024 units and ReLU activation, and a final dense layer with softmax activation for the number of classes (4 in this case).

Compilation and Training

The model is compiled with the Adam optimizer and categorical cross-entropy loss. It is then trained for 10 epochs.

Here's how the model was trained:

# Data augmentation and generators
from tensorflow.keras.preprocessing.image import ImageDataGenerator

train_datagen = ImageDataGenerator(
    rescale=1./255,
    shear_range=0.2,
    zoom_range=0.2,
    horizontal_flip=True
)

# ... (similar setup for test and validation generators)

# MobileNetV2 base model
from tensorflow.keras.applications import MobileNetV2

base_model = MobileNetV2(weights="imagenet", include_top=False, input_shape=(224, 224, 3))

# ... (freeze base_model layers and add custom classification head)

# Compilation and training
model.compile(optimizer="adam", loss="categorical_crossentropy", metrics=["accuracy"])

history = model.fit(
    train_generator,
    steps_per_epoch=train_generator.samples // train_generator.batch_size,
    validation_data=validation_generator,
    validation_steps=validation_generator.samples // validation_generator.batch_size,
    epochs=10
)

Training Results

EarthFinesse achieved remarkable training results, setting a new benchmark in terrain classification:

Final Accuracy

The model achieved a stunning final accuracy of over 97.87%, showcasing its robust performance in classifying terrain types. This high accuracy can significantly enhance the effectiveness of military operations.

Confusion Matrix

Training History

Epoch	Loss	Accuracy	Validation Loss	Validation Accuracy
0	0.274514	0.897999	0.180294	0.934242
1	0.151308	0.945084	0.208954	0.924763
2	0.121970	0.956086	0.170471	0.941647
3	0.101868	0.962776	0.154959	0.947571
4	0.090680	0.967120	0.118927	0.961345
5	0.080031	0.970640	0.128688	0.959271
6	0.073431	0.974317	0.131562	0.957198
7	0.071057	0.974508	0.123268	0.961197
8	0.064471	0.977139	0.129367	0.958235
9	0.059202	0.978661	0.114494	0.966380

These training metrics illustrate the model's progression over the training epochs, with both training and validation accuracy steadily increasing.

Applications

The EarthFinesse Military Terrain Classifier has versatile applications across various domains, including but not limited to:

1. Military Operations

• Tactical Planning: The classifier assists military strategists in understanding the terrain composition, helping them plan tactical maneuvers effectively.

• Mission Customization: Military missions can be customized based on the terrain type, optimizing resource allocation and troop deployment.

• Camouflage Strategies: Knowledge of terrain types aids in developing appropriate camouflage strategies to blend in with the surroundings.

2. Environmental Monitoring

• Conservation Efforts: Conservationists can utilize the classifier to monitor and protect specific ecosystems, such as marshlands and forests.

• Disaster Response: During natural disasters, the classifier can identify affected terrain types, aiding in disaster response and recovery efforts.

• Ecological Research: Researchers can employ the classifier for ecological studies to analyze terrain diversity and its impact on local ecosystems.

3. Agriculture and Land Management

• Precision Agriculture: Farmers can make data-driven decisions by assessing soil types and choosing optimal crops for specific terrains.

• Land Development: Urban planners and land developers can benefit from understanding the terrain for sustainable land use.

4. Autonomous Vehicles

• Navigation: Autonomous vehicles, such as drones and self-driving cars, can use terrain classification for safe and efficient navigation.

• Obstacle Avoidance: Identifying rough or impassable terrains helps autonomous vehicles avoid obstacles and hazards.

5. Geographic Information Systems (GIS)

• Map Creation: The classifier contributes to the creation of detailed maps by categorizing terrains accurately.

• Geospatial Analysis: Geospatial analysts can integrate terrain data for comprehensive geospatial analysis.

These applications demonstrate the broad utility of the EarthFinesse Military Terrain Classifier in various fields, enhancing decision-making and resource optimization.

Contributing

We welcome contributions from the community! If you'd like to contribute to this project, please review our contribution guidelines.

License

This project is licensed under the MIT License.