The image classifier consists of two interacting modules: feature extraction and linear classification. Feature extraction relies on evaluating spatial derivatives of integer or fractional order of the image, followed by non-linear transformations in the Fourier domain. Linear classification relies on multivariate statistical analysis of feature vectors. Training minimizes a loss function: the latter depends on the parameter n-tuple which controls feature extraction. Application: discrimination of bacterial spores among airborne particulate material.

This presentation is part of Contributed Presentation “CP4 - Contributed session 4”

Authors:

Giovanni Franco Crosta (University of Milan Bicocca, Dept. Earth- and Environmental Sciences)

Keywords:

environmental monitoring, machine learning, partial differential equation models