ML based Wildfire detection and early warning system

Sensor and location based system for wildfire preditions

BeginnerProtip182

ML based Wildfire detection and early warning system

Things used in this project

Hardware components

Seeed Studio Wio Terminal

Seeed Studio Grove - Air quality sensor v1.3

Grove Vision AI

Seeed Studio Grove - I2C High Accuracy Temp&Humi Sensor (SHT35)

Software apps and online services

Arduino IDE

Google Cloud IoT Core

Story

In the recent years wildfires have become a problem for many of us living on this planet. While there are technologies being used to overcome this global problem there is still a growing demand to fight the wild fire and climate change.

The Seeed studio's SenseCAP K1100 platform is suitable to fight the issue. It comes with a Grove modules like camera, Air Quality, Temprature & humidity sensors and Wio Terminal as a main interface board to these sensors.

The first part was to get started with the product. As the Wio terminal has two grove connectors for Grove modules and 40-pin raspberry pi compatible headers to plug it directly to raspberry pi, I used those 40-pin connectors to interface extra sensors to the Wio Terminal via I2C bus. From the below images it is clear that the Grove port on the left side is for I2C interface and on the right side is for UART interface.

Project building

For sensing of wild fire using ML based techniques one needs to train the image processing model which can detect wild fires effectively. For that, I used the online available datasets for fire to train and deploy the model. I followed seeed's guide to train and deploy the image processing model to vision AI kit. Up to four different models can be deployed in the device. And it can be selected from the Arduino compatible code based on the requirements of the user.

For example, the following code will select the first model loaded onto the device.

if (ai.begin(ALGO_OBJECT_DETECTION, MODEL_EXT_INDEX_1)) // Object detection and externel model 1

Let's get into some of the details of training of the model. For labeling the dataset Roboflow can be used. It is an online platform where image data can be uploaded and labeled. After that, a YOLO platform can be used to finish the training and deploying the model either online or in offline mode.

After successful completion of the model training you can download the model to the device. The following screenshot shows model-1.uf2 which can be downloaded to the Grove vision AI module by plugging in the device to the PC and pressing the boot button twice.

There are a few software design parts which needs to be integrated in my project. The first part is to integrate the sensors in one code to make it work together. You will find the example code in the attachments of this project which interfaces all the required sensors.

Sensor testing

#include <Arduino.h>
#include "Adafruit_SHT4x.h"
#include "SparkFun_SGP30_Arduino_Library.h"
#include <Wire.h>
#include <rpcWiFi.h>


SGP30 AirQ;
Adafruit_SHT4x Weather = Adafruit_SHT4x();

long t1, t2;

void setup() {
  // put your setup code here, to run once:
  Serial.begin(9600);
  Wire.begin();
  Wire.setClock(400000);
  
  while (!Serial){
    delay(10);
    }
    
  WiFi.mode(WIFI_STA);
  WiFi.disconnect();
  delay(100);
  
  
  
  if(AirQ.begin() == false){
    Serial.println("No SPG30 detected. Check connections. ");
    while(1);
    }
  AirQ.initAirQuality();
  t1 = millis();
  
  if(! Weather.begin()) {
    Serial.println("Could't find Weather sensor");
    while (1) delay(1);
    }

    Weather.setPrecision(SHT4X_HIGH_PRECISION);
    Weather.setHeater(SHT4X_NO_HEATER);
    
}

void loop() {
  // put your main code here, to run repeatedly:

  int n = WiFi.scanNetworks();
  Serial.println("scan done");

  if (n==0) {
    
    }
  else{
    Serial.println(n);
    for (int i=0; i< n; i++)
    {
      Serial.print(WiFi.SSID(i));
    }
    }
  
  
  //First AirQ measures
  t2 = millis();

  if ( t2 >= t1 + 1000)
  {
    t1 = t2;
    AirQ.measureAirQuality();

    Serial.print("CO2 ");
    Serial.print(AirQ.CO2);
    Serial.print("ppm\tTVOC: ");
    Serial.print(AirQ.TVOC);
    Serial.println(" ppb"); 
    
    //get raw values for H2 and Ethanol

     AirQ.measureRawSignals();
     Serial.print("Raw H2: ");
     Serial.print(AirQ.H2);
     Serial.print(" \tRaw Ethanol: ");
     Serial.println(AirQ.ethanol);    
    
  }

  sensors_event_t humidity, temp;
  Weather.getEvent(&humidity, &temp);

  Serial.print("Temprature:  "); Serial.print(temp.temperature); Serial.println(" degree C");
  Serial.print("Humidity:  "); Serial.print(humidity.relative_humidity); Serial.println(" % rH");

  delay(1000);

}