Domain Adaptation (DA) techniques have been applied to alleviate the domain shift problem in recent years. Mostly used in changing environments like the ones faced by self-driving cars, DA methods have greatly benefited Semantic Segmentation applications in several fields. 

Motivated by this scenario, this research aims to apply DA approaches to deforestation detection applications, hypothesizing that the performance of the classifiers will improve when new satellite images are evaluated. The research follows mainly representation matching and appearance adaptation approaches regarding DA techniques. The code and the datasets produced in this research can be found in the following link.

Deep Learning (DL), mostly based on the so-called Convolutional Neural Networks (CNN), has become the dominant trend in several Computer Vision (CV) tasks. Specifically for underwater image analysis and classification, their impact is positively significant.

In this sense, this research line aims to apply DL-based techniques to underwater image characterization from different points of view, namely, substrata and the lying fauna. The current research is being developed jointly with researchers from Ifremer and IMT-Atlantique institutions.