Technology

Air Cognizer uses Machine Learning algorithms to predict the Air Quality.

The user captures the images in an open environment from which, features are extracted and fed to the model.

AQI is determined using two Machine Learning Models:

a) Image based model

b) Meteorological parameters based model

a) Image based Model

The following feature are extracted from the captured image to predict the AQI:

1. Transmission

2. Blue Colour of the Sky

3. Gradient of Sky Region

4 .RMS Contrast and ROI Contrast

5. Entropy

6. Humidity

Transmission is the one of the major features, which affect the PM 2.5 levels.

Due to light scattering, the sky is blue on a clear day and gray or white on a hazy or cloudy day. The color of the sky is found by the average value of the blue component of RGB channels in sky region.

The smoothness of the sky is defined by the average of the gradient amplitude in the sky region. If the pollution level is high, then the gradient would be high.

Other parameters like RMS contrast and Entropy are found mathematically.Humidity is a meteorological parameter. Since, most of these parameters obey linear relation with PM 2.5, they are fed to a neural net with linear activation function.

b) Meteorological Parameters based Model

To increase the accuracy of the system , another time based model is used.

Finally, results of both the models are combined using Ada-Boosting Algorithm of Machine Learning to get an estimation of AQI.

To know more about the technical details, read our guest blog post for TensorFlow, Predicting Air Quality with TensorFlow Lite.