AIES-MAC

Automatic Information Extraction System for MACro-organisms

Automatically identifying various organisms in underwater images is challenging due to the inherent complexity of benthic images, often containing a large amount of visual elements.

ANERIS aims to create AIES-MAC - a technology to detect and identify macro-organisms in single images obtained from citizen smartphones or EMUAS, after processing by ATIRES, to support the AMOVALIH classification service by uploading observations for validation

(1) Within EMUAS, AIES-MAC refines the classification of species, the confidence in the classification, and can help detect unknown species by indicating detections for which the explainable features are very different from other specimens of the same species. It also adds new features to the existing EMUAS database by segmenting the organisms and extracting size, shape, texture, or other parameters for characterization and taxonomic classification, as well as criteria on the image quality.

(2) A specific version has been created to work on the MINKA database, to automatically classify, segment, and characterise species in the uploaded pictures. It can be retrained to segment the macro-organisms within the images, using for training only the name of the species identified (weak annotations, no need for segmentation ground truth); its performance is under evaluation.

Currently, AIES-MAC is testing preliminary algorithms using a test database of offline images & videos, with and without image restoration provided by ATIRES.