Coastal surface change detection by aerial image segmentation using convolutional neural networks

The conservation of ecosystems in coastal areas must be considered within a sustainable framework with anthropogenic activities. It is relevant to quantify the impact generated by external agents, hence the objective of this work is to implement a method for monitoring the degradation of the beach s...

Descripción completa

Detalles Bibliográficos
Autores: Padilla-Arballo, Jacquelina J., Martínez-Díaz, Saúl, Castro-Liera, Marco A., Luna-Taylor, Jorge E.
Tipo de recurso: artículo
Estado:Versión publicada
Fecha de publicación:2022
País:México
Institución:UNIVERSIDAD AUTÓNOMA DEL ESTADO DE HIDALGO
Repositorio:PÄDI Boletín Científico de Ciencias Básicas e Ingeniería del ICBI
Idioma:español
OAI Identifier:oai:repository.uaeh.edu.mx:article/9290
Acceso en línea:https://repository.uaeh.edu.mx/revistas/index.php/icbi/article/view/9290
Access Level:acceso abierto
Palabra clave:Convolutional neural networks
Semantic segmentation
Time series
Protected coastal areas
Redes neuronales convolucionales
Segmentación semántica
Series temporales
Zonas costeras protegidas
Descripción
Sumario:The conservation of ecosystems in coastal areas must be considered within a sustainable framework with anthropogenic activities. It is relevant to quantify the impact generated by external agents, hence the objective of this work is to implement a method for monitoring the degradation of the beach surface and adjacent coastal vegetation. Captures of aerial images of protected coastal areas have been collected, obtained periodically by means of a drone vehicle. A dataset was integrated that includes all seasonal phases and distinguishes 5 classes for monitoring: Mangrove (mangle), creeping vegetation (vegetación rastrera), sand (arena), sea (mar) and hill-plain (cerro-planicie). For semantic segmentation different convolutional neural network (CNN) architectures were implemented and compared using transfer learning. The results have been robust in the classification, reaching an overall accuracy of 93.9% and between 89.9-95.8% in individual classes. In the Intersection over Union, IoU metric, the range was between 86.6-92.7%. In change detection, time series are used for class monitoring. This method has been applied to the case study of Ensenada Grande beach in the Parque Nacional Archipiélago Espíritu Santo.