GeoPeak-Finder: Terrain Digital Twin & Optimization

Terrain Digital Twin and Geographic Peak Optimization Identification

---

English Version | 中文版本

---

English Version

Live Demo

• Public Link: https://geopeak-finder-qumxnbzjdabafzeyjla6cs.streamlit.app/

Project Overview GeoPeak-Finder is an AI-driven geospatial project designed to create a Terrain Digital Twin of the Pittsburgh area. By fitting discrete elevation data with PyTorch Neural Networks, the system generates a differentiable surface used for automated peak identification via Multi-start Gradient Ascent (SGA) and advanced environmental risk analysis.

Project Structure

GeoPeak-Finder/
├── data/               # Raw geospatial elevation data
├── models/             # Saved model weights (.pth)
├── results/            # Generated analytical outputs
│   ├── 1_basic_gradient_map.png
│   ├── 2_flood_risk_analysis.png
│   └── 3_landslide_risk_analysis.png
├── src/                # Core implementation logic
│   ├── model.py        # PyTorch Neural Network architecture for terrain fitting
│   ├── engine.py       # Training pipelines and SGA optimization algorithms
│   ├── utils.py        # Data loaders and geospatial preprocessing tools
│   └── visualizer.py   # Plotting logic for terrain, risk maps, and trajectories
├── app.py              # Streamlit-based interactive web interface
├── main.py             # Main entry point for training and optimization
└── requirements.txt    # Project dependencies

Installation & Usage

1. Setup:

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2. Run Pipeline:

python3 main.py

3. Interactive UI:

streamlit run app.py

Core Results:

• Peak Identified: Robert Williams Reservoir (~365m).
• Flood Risk: Automated generation of flood susceptibility maps based on neural terrain gradients and flow accumulation.
• Landslide Risk: Automated generation of landslide susceptibility maps based on neural terrain slope and gradient analysis.

---

中文版本

GeoPeak-Finder: 地形数字孪生与寻优

基于深度学习的地形建模与多起点梯度上升算法实现

---

在线演示

• 链接: https://geopeak-finder-qumxnbzjdabafzeyjla6cs.streamlit.app/

项目简介 项目概览 GeoPeak-Finder 是一个人工智能驱动的地理空间项目，旨在构建匹兹堡地区的地形数字孪生（Digital Twin）。通过利用 PyTorch 神经网络拟合离散的高程数据，系统生成了一个可微的曲面，用于通过多起点梯度上升算法（SGA）进行自动峰值识别和高级环境风险分析。

项目结构

GeoPeak-Finder/
├── data/               # 原始地理空间高程数据
├── models/             # 已保存的模型权重 (.pth)
├── results/            # 生成的分析结果输出
│   ├── 1_basic_gradient_map.png
│   ├── 2_flood_risk_analysis.png
│   └── 3_landslide_risk_analysis.png
├── src/                # 核心实现逻辑
│   ├── model.py        # 用于地形拟合的 PyTorch 神经网络架构
│   ├── engine.py       # 训练流水线与 SGA 寻优算法
│   ├── utils.py        # 数据加载器与地理空间预处理工具
│   └── visualizer.py   # 地形、风险图及寻优轨迹的绘图逻辑
├── app.py              # 基于 Streamlit 的交互式 Web 界面
├── main.py             # 训练与寻优流水线的主入口
└── requirements.txt    # 项目依赖项

安装与使用

1. 环境初始化：

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2. 运行pipeline：

python3 main.py

3. 交互式界面：

streamlit run app.py

核心成果：

• 识别的峰值: Robert Williams Reservoir (罗伯特·威廉姆斯水库，约 365 米)。
• 洪涝风险: 基于神经地形梯度与汇水分析，自动生成洪涝易发性风险图。
• 滑坡风险: 基于神经地形坡度与梯度分析，自动生成滑坡灾害易发性风险图。