Innovative Process for the Geo-linked Analysis of Air Quality Assisted by Ai Techniques

Image recognition is a technology that, through specific algorithms and methodologies, aims to reproduce the typical biological vision systems by identifying particular objects, patterns, colors or geometric shapes. Artificial intelligence was used in this study to define and test an algorithm for image recognition. In recent years it has been applied to the pollutants pattern recognition to evaluate some important phenomena and to catch details in the pollutants dispersion that with the standard measurement systems it is not possible. This study therefore aims to train an artificial intelligence in the automatic recognition of the air pollution level measured in the city of Milan obtained by a new generation sensor network. Air quality data came from the real-time on-road monitoring network operating in Milan. The developed algorithm has been applied to the PM10 concentration maps that divides the area in cells with an area of 1 km2. The pollutant concentrations are reported on the map according a color scale starting from green to yellow and red according to the law limit value according D.Lgs. 155/2010. So, the color scale defines three levels – good, medium and bad- in which the dataset has been divided. Subsequently, the map where divided into four macro-areas (North West, North East, South West, South East) and the algorithm has been trained to identify the pollution patterns related to each of the four macro-areas. Sensitivity analysis were carried out on the model hyper-parameters to test the model and to increase its performance. In this work, the computer system designed to analyze, process and recognize the images provided by Google, Teachable Machine, was used, which through the use of artificial intelligence, allows the automatic images recognition.