Trailblazers| Spot That Fire V3.0
Spot That Fire V3.0
Spot-That-Fire-V3.0-Nasa-Space-Apps-Challenge
Summary
Our project leverages pioneering AI techniques to detect wildfire at very early stages, way faster than the current systems used. The system seamlessly detects fire with data from various sources like IoT devices, satellite data and data provided by users living nearby the forest. The use of different technologies gives us a unique advantage and helps detect wildfire faster. Our system also provides great features like smart routing, smart alert system, emergency centre finder, and much more to the users in case a wildfire occurs.As soon as the Fire is detected we will inform the nearest Fire stations and emergency centres in particular radius by Email, SMS, or any other Push notifications.
How We Addressed This Challenge
We developed WebApp and Mobile App where we keep records of all Fire stations and Emergency Centers with Residents living near Forest as Users or anyone interested. Each record will store their location coordinates for Geo Location mapping.
Detection by Users Uploaded Image
Users can register, and their current location will be used to predict the danger and will also suggest the contacts of nearest Fire Stations, Rescue Centres. Users can view the previous Wildfires, report one by clicking a picture, along with their location, which will be verified by other users on their application, the satellite images and the Image Classifier (Deep Learning Model) at back-end. On registration, user will be added to the nearest Rescue Centre’s Discussion Group where the important information and notice will be provided.
Report Fire Web Page
Detection by IoT Devices installed near forest or around forest
We will collect the data through these devices on periodic basis and the data includes attributes like Oxygen Level, Humidity, Temperature, CO2, CO levels to predict or detect fire chances in the forest. Our Machine learning model will predict and server will issue Alert through push notifications to all nearest fire stations and users/residents near the device location.
Detection by Satellite Images
We would also add some cron jobs to check images of major or frequent forests that catches fire through our deep learning model. This can also be used to verify above 2 methods also if Live satellite images can be used.
Users as well as the IoT devices can report the WildFire using the API, which will be verified by Deep Learning Model at back-end, Satellite Images and marked valid/invalid by users. On Successfully detecting the Fire, the nearest Fire Stations will be alerted, users in a particular radius will be sent push notifications and alerts simultaneously, Rescue Centres will be notified to prepare with proper supplies and users will be directed with a safe path to rescue centres.
Users can View nearest Fire Stations or Emergency Centre's as well as they can also see nearby Fire Alerts in App also (user's will receive a Push Notification) at the time of Detection of fire.
Nearby Rescue Centers or Emergency Centers
Nearby Fire Stations
Recent Fire Reports
How We Developed This Project
Currently all the solutions take minimum 3-4 hours to Detect and Validate the Forest Fire. This makes fire to spread and which becomes more dangerous and harder to get control. Therefore, wanted to make something which can Detect, Validate and Alert fire in not more than 15-20 minutes. This was a difficult task to implement all Deep Learning and Machine Learning models with API and Other information secondly we also added the role of User/Residents living nearby by just uploading 1 picture they can alert all nearby Users/Concerned Depts.
Web Server
We used Django as Backend Framework as it's a Python Framework therefore integrating efforts and response time is reduced. We used Geo Location for Distance and Push Notifications and Celery - Redis for Background Tasks.
Deep learning model :
For Image classification purpose we used deep learning technique called CNN (Convolutional Neural Network). We used custom CNN architecture. The model uses 150 x 150 size images as input for classification.
We used TensorFlow, NumPy, matplotlib and h5py for the deep learning model. We used the FIRE Dataset for training our deep learning model.
The accuracy and loss of our model is shown below (Accuracy = 97.16%, Validation_accuracy = 98.30%, Loss = 0.0664%, Validation_loss = 0.0476%).
We used FIRE Dataset (https://www.kaggle.com/phylake1337/fire-dataset) from Kaggle for training the image classification deep learning model.
Machine Learning Model :
Users as well as the IoT devices can report the WildFire using the API, which will be verified by Deep Learning Model at back-end.
We created a custom Dataset (CSV) containing columns like temperature value, humidity value, oxygen value.
We're successfully able to detect fires through both the given methods and was able to send Email to all Nearby Fire Stations, Rescue Stations, Users/Residents on the basis of location of fire.
How We Used Space Agency Data in This Project
We will use satellite Images to detect real time Fires through Deep Learning Model. It can be useful to Authenticate the requests received by our Both the methods discussed above.
Project Demo
PPT
https://drive.google.com/file/d/1ydoTnrA7q-vr-CmIoFrXm0KEO2VmNRjz/view?usp=sharing
Project Code
Data & Resources
References :
Deep learning model :
We used FIRE Dataset(https://www.kaggle.com/phylake1337/fire-dataset) from Kaggle for training the image classification deep learning model.
Machine Learning Model :
We created a custom Dataset (CSV) containing columns like temperature value, humidity value, oxygen value.