Antriksh - Space Apps Challenge

Antriksh| Automated Detection of Hazards

Awards & Nominations

Antriksh has received the following awards and nominations. Way to go!

Global Nominee

Automated Detection of Hazards

Countless phenomena such as floods, fires, and algae blooms routinely impact ecosystems, economies, and human safety. Your challenge is to use satellite data to create a machine learning model that detects a specific phenomenon and build an interface that not only displays the detected phenomenon, but also layers it alongside ancillary data to help researchers and decision-makers better understand its impacts and scope.

Automatic Dust Storm Detector(ADSD)

Summary

Dust Storms are nimble beasts of nature that destroy more than they restore, and our project helps to identify them in real time. It uses Stacking Ensemble using Deep Neural Networks to train satellite images further tested on data acquired on our web interface from GIBS API in real time to predict whether or not the input image is a dust storm. Layered data such as aerosol index and dust score are considered. The image is also passed through a K-Means segmentation model to estimate the areas affected by the dust storm. NLP is used to extract possible mentions of locations from social media and predict dust storms in the requested area, and notifies in case it does predict there is one.

How We Addressed This Challenge

What did you develop?

We have developed an application that gives real time predictions of satellite data to predict the availability of dust storms in the image. This is followed by a depiction of the area affected by the dust storm. In case of a dust storm being detected, a notification is sent to the users. It also uses social media as a tool to extract possible occurrences of dust storms and predict the possibility of it being there, considering the location can be extracted.

Why is it important?

Dust storms have been wreaking havoc amongst humanity as well as flora and fauna by destroying crops, being the cause of aggravation in respiratory health issues and decrease of the overall Air Quality Index (AQI) of areas affected.

Our application informs its users about dust storms occurring in real time, so they can be aware of any possible dust storm hazard in their locality and stay alert virtually. This is especially crucial for drivers who might be heading towards a dust storm, and could possibly avoid it. It also could alert people with respiratory ailments to take the necessary precautions in time.

What does it do?

It helps in detecting dust storms which are going to hit a particular region of the world by doing early prediction through classifying them against Aerosol,Dust Score and Corrected reflectance ,after which it applies segmentation on the predicted data to know exact area of impact which is compared against population and vegetation data of the region. It also sends a warning of hazard and holds historical prediction data.

How does it work?

Image data is initially collected from the Global Imagery Browse Services (GIBS) archive using Worldview. In the image data each sample set is considered as a set of three images with the respective layers::

Corrected Reflectance layer (True Color) - VIIRS Instrument, Suomi NPP satellite
Aerosol index image layer - OMI Instrument, Aura Satellite
Dust Score Data layer - AIRS Instrument, Aqua Satellite

These images then go through general preprocessing functions followed by Data augmentation using DataAugment library which uses Bayesian Optimization to optimize the data augmentation hyperparameters.

The images are then trained

These processed images are then trained on Stacking Ensemble which utilizes neural networks to classify the images into whether they are a dust storm or not.

Images with the respective layers are now retrieved in real time using the GIBS API and those images are passed through the trained ensemble architecture to determine whether the image has a dust storm in it or not.

The image retrieved is then passed through a semantic segmentation algorithm using K-Means to segment the data so as to determine what area of land has been affected by the dust storm.

Impact analysis is implemented where against survey data available, the population based on the population density of the retrieved area, vegetation and settlement is displayed with the help of worldwide mapbox API, which gives a peek into the possible after effects of the storm.

Finally, the application alerts the users by notifying them of the hazard in the place as well as analyzing disaster related news and tweets using NLP fetching.

What do you hope to achieve?

We are planning to implement localization to pinpoint the location of the dust storm, as well as use the real time data resources to predict where the dust storm is headed next based on parameters such as wind speed and wind direction. We also hope to expand the implemented architecture into other parameters such as precipitation, radiation as well as look further into how this affects biodiversity such as agriculture through thorough studies of soil moisture and carbon emission impacts of dust storms.

Finally, we want to make this platform as efficient as possible so people everywhere need not have to worry about their safety in the event of dust storms.

How We Developed This Project

We had a team of 6, where 2 of us worked mainly on searching and collecting the dataset, and deciding on the machine learning algorithm to be applied. Another 2 worked on the front end to create the website as well as incorporate respective features. Then we had 2 team members who were looking into the documentation and description of the project as well as implementations of certain codes from the frontend, backend as well as the machine learning models, if required. It was an overall collaborative teamwork in a virtual framework, and it turned out to be well coordinated and synced.

Dust Storms look relatively harmless as compared to some other hazards as they don’t tend to destroy buildings and monuments, but what people forget to consider is the damage it does to ecology and overall health of a specific area. Dust particles are silent killers who cause respiratory diseases, possibly fatal heart and lung damage as well as transmit diseases by travelling through air. This makes them equally, if not possibly more dangerous than many other hazards, and as the challenge demands, to inform people about the presence of this deadly hazard seemed to be a crucial task which had to be taken in hand.

We want to make the world a more alert and safer place, one hazard at a time.

We faced problems regarding data collection, as a lot of time went in deciding the hazard and downloading data. The model was then selected in which selecting the ensemble method to choose, which took some time as well. We were able to augment and get real time data vert easily, thanks to the GIBS API.