Regression modeing allows us to estimate numerical quality parameters from spectral data. For example, we may wish to estimate sugar content in a tomato or mixing proportion of powders. In perClass Mira, regression is performed at object level. Therefore, we need to define pixel classifier and one or more classes of interest. Then, we can assign external numerical values to each object and build a regression model. This model is then applicable to objects detected in a new image and can provide e.g. an estimate of sugar content per tomato.


In this example, we use the powder data set with vials containing mixtures of two powders, namely flower and soda. Our goal is to train a model that will be able to estimate the mixing propotion for a new powder mix.


We have loaded the first scan with powders: