Autor(es): Gomes, João Pedro ; Sousa, Joaquim J. ; Pádua, Luís ; Cunha, Carlos R. ; Cunha, António
Data: 2021
Identificador Persistente: http://hdl.handle.net/10198/27306
Origem: Biblioteca Digital do IPB
Assunto(s): Seaweed; Macroalgae; Hyperspectral; UAV; Deep learning; Machine Learning
Macroalgal communities are generally found in coastal regions, close to rocks or other hard surfaces. They provide shelter and food for many organisms and are of interest to the food industry, pharmaceutical, and agriculture. They are also an indicator of environmental change. Traditionally, the process of identification and monitoring of these communities and their constituent species is based on manual methods. The use of unmanned aerial vehicles (UAVs) allows the remote collection of images with high spatial and spectral resolution and adjustable time scales. The development of methodologies allowing the processing and the analysis of UAV-based high-resolution imagery would be of economic and environmental importance. That would allow to streamline the identification of species with economic potential, the evaluation of the seasonal and spatial variation of the available biomass, and the monitoring of the coastal ecological status and its evolution. Recent technological developments in the areas of remote sensing and artificial intelligence make it possible to provide tools with great potential for these applications. Indeed, hyperspectral sensors can nowadays be coupled in UAVs allowing for high spatial and spectral resolution imagery. The data processing powered by deep learning and its increasing diversity of models and architectures is the ideal way to handle and analyze the huge volume of data acquired. In this paper, we describe a methodology to be implemented in a system to be developed to make the automatic classification of existing species in macroalgal communities, using deep learning models applied to hyperspectral images collected by UAVs.
