Análisis del blanqueamiento coralino en fotografías subacuáticas con redes neuronales artificiales

Abstract

Coral bleaching, a phenomenon that affects coral reefs around the world, is a crucial indicator of the health of marine ecosystems. The ability to accurately monitor and evaluate bleaching is critical to understanding its impact and taking effective conservation measures. The present work addresses the challenge of coral bleaching by developing a methodology based on artificial neural networks to analyze underwater photographs and quantify the degree of coral bleaching. The methodology consists of three phases: detection of corals for which the Grounding DINO model was used, segmentation of corals with the Segment Anything model and quantification of bleaching with the K-means algorithm. To validate it, an image bank was used and the results were compared with manual segmentation carried out with the help of a coral expert. High efficiency was demostrated in the detection and segmentation of corals, as well as in the identification of areas affected by bleaching in most cases. However, challenges related to color quantization were identified, especially under lighting conditions and specific coral characteristics. This study represents a significant advance in the development of automated tools for monitoring coral bleaching, offering new perspectives for the research and conservation of marine ecosystems.