Weighted Hausdorff Distance Loss as a Function of Different Metrics in Convolutional Neural Networks for Ladybird Beetle Detection

Producción científica: Capítulo del libro/informe/acta de congresoContribución a la conferenciarevisión exhaustiva

Resumen

This work compares five different distance metrics (i.e., Euclidean, Chebyshev, Manhattan, Mahalanobis, and Canberra) implemented in the weighted Hausdorff distance (WHD) as part of the loss function during the training and validation of a fully convolutional neural network (FCNN) model for detecting ladybird beetle specimens. The FCNN-based detector was trained and validated using a ten-fold cross-validation method on a database composed of 2,633 wildlife images with ladybird beetles. The obtained results highlighted the Chebyshev metric as the top performer given a diverse dataset as ours. This metric scored the highest values in three out of four validation metrics (i.e., precision, recall, and F1-score). The nature of this metric allows substantial space for minimizing the cost function along the FCNN training step. Euclidean and Manhattan distances also provide good performance based on our validation metrics, while Mahalanobis and Canberra distances are not suitable for detecting of ladybird beetles.

Idioma originalInglés
Título de la publicación alojadaApplications of Computational Intelligence - 4th IEEE Colombian Conference, ColCACI 2021, Revised Selected Papers
EditoresAlvaro David Orjuela-Cañón, Jesus A. Lopez, Julián David Arias-Londoño, Juan Carlos Figueroa-García
EditorialSpringer Science and Business Media Deutschland GmbH
Páginas65-77
Número de páginas13
ISBN (versión impresa)9783030913076
DOI
EstadoPublicada - 2022
Evento4th IEEE Colombian Conference on Applications of Computational Intelligence, ColCACI 2021 - Virtual, Online
Duración: 27 may. 202128 may. 2021

Serie de la publicación

NombreCommunications in Computer and Information Science
Volumen1471 CCIS
ISSN (versión impresa)1865-0929
ISSN (versión digital)1865-0937

Conferencia

Conferencia4th IEEE Colombian Conference on Applications of Computational Intelligence, ColCACI 2021
CiudadVirtual, Online
Período27/05/2128/05/21

Huella

Profundice en los temas de investigación de 'Weighted Hausdorff Distance Loss as a Function of Different Metrics in Convolutional Neural Networks for Ladybird Beetle Detection'. En conjunto forman una huella única.

Citar esto