Team NA's| Automated Detection of Hazards
Automated Detection of Hazards
Air Quality Regressor for Automated Air Pollution Detection
Summary
Our project focusing on detecting air pollution by evaluating the PM 2.5 value. Using deep learning, our regressor model can estimate the air quality in Particulate Matter 2.5 (PM2.5) measurement provided a few other factors which are Relative Humidity (RH), UGRD & VGRD, Height of Planetary Boundary Layer (HPBL), Temperature (TMP) and Aerosol Optical Depth (AOD )reading from GOES-R. The value is then evaluated to determine the quality of air.
How We Addressed This Challenge
What did you develop?
Throughout the challenge we develop a neural network model to predict the PM2.5 that will be evaluated to determine the quality of air. We also build a web application that serves as an interface for user to enter the information related and use our trained model to evaluate the air quality.
Why is it important?
Developing a deep neural network to predict the air quality is the fundamentals to solve the challenge which is to create an automated hazards detection. Air pollution is all around us. Indoors, outdoors, in cities and in the countryside. It affects us all, whether we realize it or not. Recent research has started to highlight some rather worrisome aspects of the component of the air around us really contains and it really affects our body. Without air, life can be endangered as breathing polluted air condemns us to a life of disease and early death. Currently, the detection of quality of air is done manually by a human however, there is possibility of human errors during detection which can lead to a false alarm and cause unnecessary panic. Moreover, time consumption while detecting can be reduced.
What does it do?
Our web application can be used by anyone to predict the value of PM2.5. PM2.5 refers to atmospheric particulate matter (PM) that have a diameter of fewer than 2.5 micrometers.
How does it work?
Resources provided by NASA is used to develop a deep neural network model. The model is trained using a few related information such as RH, UGRD & VGRD, HPBL, temperature and AOD reading from GOES-R. The trained model is then deployed into the web application which can be accessed and used by anyone.
What do you hope to achieve?
At the end of this challenge, we hope that we can learn a lot about how to manage and utilize the resources provided. We hope that the system can be a used by everybody to predict the quality of air every day for their awareness so that the can prepare if there is bad quality of air that they are facing. We also planned to improve our neural network model and add more functionalities in the web application such as an alert message to notify users.
How We Developed This Project
What inspired your team to choose this challenge?
We want to apply what we learned in our studies regarding Deep Learning and Artificial Intelligence.
What was your approach to developing this project?
We are focusing on implementing a deep neural network model to train the datasets. Naturally, training a deep learning model is highly expensive, it requires a lot of time. Thus, we only build a model prototype for this hackhaton because of time restriction.
What tools, coding languages, hardware, software did you use to develop your project?
We are using Keras, TensorFlow library to design and train the model. The main environment we are using is Google Colaboratory. Then, We use the model and deploy it in a web app which hosted in heroku. The language which widely used in this project is Python.
What problems and achievements did your team have?
We managed to build the prototype model and a web application : https://air-quality-regressor.herokuapp.com/
How We Used Space Agency Data in This Project
In this project, we used the labeled data provided in NASA github repository. The data contain the information of PM2.5 readings collected by EPA from various stations. The CSV formatted data and labels are provided below. It contains the following fields which have been described.
- station_id: Unique identifier of the PM 2.5 monitors stationed across US
- stime: Time and date of sample recorded
- air_data_value: EPA air data PM2.5 readings
- RH: relative humidity from HRRR
- UGRD, VGRD: Wind speed vectors from HRRR
- HPBL: Height of Planetary Boundary Layer from HRRR
- TMP: Temperature recorded from HRRR
- goes_measurement: AOD reading from GOES R
Using the above information, we developed a neural network model that can predict the PM2.5 readings that will be deployed to our web application to be utilised by public.