Classification of the urban green: a methodological proposal
As part of the development activities of ISPRA National Environmental Information System, a new methodology was developed based on remote sensing images. It allows to classify the amount and type of vegetation in urban areas.
It 'was then launched the project "Urban Green" which aims to identify, at 1:5,000 scale and minimum mapping unit of 16 m2, the urban green areas of the 24 Italian provincial capitals with a population exceeding 150,000, using high resolution multispectral satellite images. The methodology includes the use of artificial neural networks and is therefore considered innovative.
Neural networks are born from the attempt to simulate the functions and capabilities of the human brain, giving the device a certain decision-making capacity due to its 'artificial intelligence'.
They consist of a large number of independent units (neurons) that form an input layer, one or more hidden layers and an output layer. Each unit is connected to other units, which are activated by a stimulus (input) of sufficient intensity and send signals to the connected equipment.
Through a 'training' process the network learns how to connect an output with an input thorough the presentation of correct examples of pairs of input / output. When training process is completed, the network is ready to classify the image. This unconventional approach is valuable in that once the network is trained for an image with certain spectrum features, it can be applied to other images with similar characteristics, thus reducing the processing time.
This paper describes the methodology applied to the project and discusses two case studies.