Multi-sensor post classification fusion
iota2 can use image time series of several sensors and perform a multi-modal classification using "early fusion" or "feature fusion", that is all data and the extracted features are stacked together and a single classification is performed.
In some cases (mainly for dimensionality purposes), it may be more interesting to perform a classification fusion. In this case, the data from the different sensors can be used independently to perform a classification (in a lower dimensionality space) and the classifications can be fused afterwards.
The fusion can be done by majority voting, Dempster-Shaffer's rule (using the confusion matrices of the classifications), confidence voting (using the confidence maps of each classification) or probability voting (if each classifier outputs the probability of each class).