🏛️ AI-Driven Archaeological Site Mapping

---

📖 Overview

AI-Driven Archaeological Site Mapping is a research-oriented computer vision project that explores how deep learning and environmental analysis can assist in identifying potential archaeological sites.

The system analyzes vegetation patterns and soil characteristics from images using AI models to detect anomalies that may indicate buried structures, ancient settlements, or historical land disturbances.

The project integrates:

• 🌱 Vegetation Segmentation
• 🪨 Soil Pattern Detection
• 📊 Visual Model Analysis
• 🧠 Explainable AI using Grad-CAM

---

🎯 Motivation

Archaeologists often rely on environmental signals such as:

• Abnormal vegetation growth
• Soil discoloration
• Surface texture changes
• Disturbed land patterns

These indicators can reveal hidden structures underground.

This project investigates how AI models can automatically detect these signals, helping archaeologists narrow down potential excavation locations.

---

🧠 Model Architecture

The deep learning system is built around a YOLO-based object detection and segmentation pipeline.

Input Image
     │
     ▼
Image Preprocessing
(resizing, normalization)
     │
     ▼
Deep Learning Model
(YOLO Segmentation / Detection)
     │
     ▼
Feature Extraction
     │
     ▼
Prediction Layer
     │
     ▼
Output
 ├── Vegetation Segmentation
 ├── Soil Classification
 └── Bounding Box Detection

---

⚙️ Project Pipeline

The complete workflow of the system:

Satellite / Ground Images
            │
            ▼
     Data Collection
            │
            ▼
     Data Annotation
            │
            ▼
      Model Training
 (Vegetation + Soil Models)
            │
            ▼
      Model Evaluation
   (F1 Curves & Metrics)
            │
            ▼
     Prediction & Testing
            │
            ▼
 Visualization & Explainability
   ├─ Bounding Box Detection
   ├─ Grad-CAM Heatmaps
   └─ Performance Curves
            │
            ▼
 Archaeological Pattern Analysis

---

📂 Repository Structure

AIDriven-Archaeological-Site-Mapping
│
├── ArchilogicalMapping/
│
├── SoilDetection/
│   ├── SoilDetection.ipynb
│   ├── SOIL.md
│   ├── bbox_visualization.png
│   ├── distribution.png
│   ├── gad_cam.png
│   ├── class_labels.json
│   └── test_soil.jpg
│
├── UI-Demo/
│
├── VegetationSegmentation.ipynb
├── VEGETATION.md
├── best.pt
├── data.yaml
├── results.csv
├── veg_test.jpg
│
├── BoxF1_curve.png
├── MaskF1_curve.png
│
├── AgroSensi-AI-2.pptx
└── README.md

---

🌱 Vegetation Segmentation

Vegetation anomalies can indicate underground structures.

This module trains a deep learning segmentation model to identify vegetation patterns.

Files

VegetationSegmentation.ipynb
VEGETATION.md
best.pt
data.yaml

Outputs

• Vegetation masks
• Bounding box F1 score curve
• Mask segmentation accuracy

Example evaluation plots:

BoxF1_curve.png
MaskF1_curve.png

---

🪨 Soil Detection

Soil composition differences often reveal hidden archaeological features.

The soil detection module performs:

• Soil classification
• Bounding box detection
• Model explainability

Files

SoilDetection.ipynb
SOIL.md
class_labels.json

Visual Outputs

• Dataset distribution plot
• Bounding box visualization
• Grad-CAM interpretability heatmap

distribution.png
bbox_visualization.png
gad_cam.png

Grad-CAM highlights which image regions influenced model predictions.

---

💻 UI Demonstration

The UI-Demo directory shows how the AI models could be integrated into a visual interface for archaeologists or researchers.

This allows easier interaction with prediction outputs and visualizations.

---

📊 Results

The trained models produce:

• Vegetation segmentation maps
• Soil classification predictions
• Bounding box detections
• Performance metrics
• Grad-CAM explanation maps

Results are saved in:

results.csv

---

🚀 How to Run the Project

Clone Repository

git clone https://github.com/ShubhamS2005/AIDriven-Archaeological-Site-Mapping.git

Move Into Directory

cd AIDriven-Archaeological-Site-Mapping

Run Notebooks

Open and run:

VegetationSegmentation.ipynb
SoilDetection/SoilDetection.ipynb

Execute cells sequentially for training and predictions.

---

🔬 Applications

This system can support:

• Archaeological site prediction
• Environmental anomaly detection
• Remote sensing analysis
• Cultural heritage preservation
• AI-assisted archaeological surveys

---

🔮 Future Improvements

Potential future extensions:

• Satellite imagery integration
• Multi-spectral remote sensing analysis
• GIS mapping integration
• Web dashboard for visualization
• Real-time site prediction system

---

Demo Link

https://archilogicalmappingui-ghgvrpkd29qhwrgmcerkyo.streamlit.app/

Demo use

Admin, pass 1234

📜 License

This project is open-source and available under the MIT License.

---

👨‍💻 Author

Made with ❤️ by Shubham Srivastava (shubhamsrivastava12568@gmail.com)

⭐ If you find this project useful, consider giving it a star on GitHub!

---