ABSTRACT
Data classification is one of the main challenges in data mining and
knowledge discovery. It is widely explored in multiple applications,
such as identifying microorganisms by analyzing images of cultures
grown in Petri dishes. This work proposes to chronologically organize
images of bacteria on solid media, captured at equidistant intervals
over several days, treating them as video frames to improve the
recognition of these microorganisms. To this end, we develop an approach
that chains classification models, integrating spatial features
from pre-trained convolutional neural networks with temporal information
propagated through target meta-features, i.e., classifier
outputs.We experimentally compared our proposal with two intentionally
designed baseline methods using a dataset with 240 images,
48 per class, and considering the macro-averaged F1 score. Results
demonstrate that addressing chronological relations enhances the
identification models’ performance, even though baseline strategies
may benefit from fewer examples.