Rancang Bangun Sistem Klasifikasi Status Bahaya Kebakaran Hutan Berbasis Edge Computing pada Jaringan Komunikasi LoRa Mesh

A full-stack Internet of Things (IoT) and TinyML project designed for early forest fire detection. This system utilizes low-power edge nodes running machine learning models locally, communicates via a robust wireless LoRa Mesh network utilizing LoRaMesher Library, and visualizes real-time environmental data through an interactive web dashboard using streamlit.

Project Structure

This repository is organized as a mono-repo containing three main architectural components:

C:.
├───dashboard-streamlit/     <-- Web-based monitoring dashboard (Streamlit)
│       ├── app.py
│       ├── config_nodes.json
│       ├── requirements.txt
│       └── serial_log.txt
│
├───firmware/                <-- C++ Coded Hardware Firmware (PlatformIO)
│   │   .gitignore
│   │   platformio.ini
│   └───src
│       ├───edge-node/       <-- Local inference & Sensor Acquisition
│       └───sink-node/       <-- Central Gateway Router
│
└───model-development/       <-- Machine Learning research & TinyML Export
        ├── MLResearch_Skripsi.ipynb
        ├── model_data.h
        └── requirements.txt

Features & Components

1. Firmware (firmware/)

Developed in C++ using PlatformIO and FreeRTOS scheduling on the ESP32 DevKit V1 platform. It operates in multiple environments via platformio.ini targets:

• Edge Node: Acquires atmospheric metrics via BME280 (Temperature, Humidity, Pressure) and air quality via MQ135 gas sensor. Performs localized TinyML classification and broadcasts hazard status via LoRa SX1278 (Ra-02) utilizing the LoRaMesher routing protocol.
• Sink Node / Gateway: Listens to incoming multi-hop mesh messages from the field network and relays packet data directly to the monitoring station over Serial communication.
• Sensor Calibration Tool: Separate standalone executable (calibrate_mq135.cpp) to precisely determine individual sensor baseline resistance ($R_0$) values before deployment.

2. Model Development (model-development/)

Contains the end-to-end TinyML pipeline implemented inside Jupyter Notebook (.ipynb):

• Features preprocessing including Z-Score Normalization and MinMax Scaling.
• Comparative study of three classification algorithms: Random Forest, Logistic Regression, and Artificial Neural Networks (ANN).
• Automated model compilation into an optimized, bare-metal compatible C-array file (model_data.h) for deployment onto constrained embedded systems.

3. Monitoring Dashboard (dashboard-streamlit/)

A responsive telemetry web interface built on Python Streamlit:

• Parses structural network definitions via config_nodes.json for dynamic node location mapping.
• Renders real-time telemetry updates from the serial pipeline.
• Features modular visualizations tracking both instant microclimate trends and systemic telemetry logs safely stored in serial_log.txt (local-only cache).

---

Installation & Build Guide

Firmware Deployment (PlatformIO)

1. Open the /firmware directory inside VS Code with the PlatformIO extension active.
2. Select your desired environment from the PlatformIO Env bar:

• env:calibrate-mq135 (For initial sensor baseline profiling)
• env:edge-node (For deployment on edge sensor nodes)
• env:sink-node (For building the gateway router)

3. Connect your target ESP32 module and execute the Build & Upload command.

Running the Web Dashboard

1. Navigate to the dashboard directory:

cd dashboard-streamlit

2. Install dependencies:

pip install -r requirements.txt

3. Boot the application:

streamlit run app.py

---

Hardware Pinout Reference (Quick Guide)

• BME280 Connection (I2C): SDA $\rightarrow$ GPIO21, SCL $\rightarrow$ GPIO22
• LoRa Ra-02 Connection (SPI): SCK $\rightarrow$ GPIO18, MISO $\rightarrow$ GPIO19, MOSI $\rightarrow$ GPIO23, NSS/CS $\rightarrow$ GPIO5, DIO0 $\rightarrow$ GPIO2
• Power Delivery Architecture: Regulated via an MT3608 2A Boost Converter fed by standard LG 18650 Li-ion batteries with embedded bulk decoupling capacitors (Panasonic FR 1000µF 16V Low-ESR) placed parallel to the VCC rail to filter out transmit-induced voltage drops.

---

Citations & Referencing

If you want to reference this source code repository inside your academic paper or undergraduate thesis, please adapt the manual bibliography details as follows:

• Type: Web Page
• Author: Paramaditya, Thoriq Kusuma
• Title: Source Code: Rancang Bangun Sistem Klasifikasi Status Bahaya Kebakaran Hutan Berbasis Edge Computing pada Jaringan Komunikasi LoRa Mesh
• Publisher/Repository: GitHub Repository
• Year: 2026
• URL: https://github.com/prmditya/wildfire-detection-system

---

License

This project is documented and open-sourced under the terms of the MIT License.