Desarrollo de un método basado en visión por computador para segmentar imágenes de los anillos de crecimiento en la especie Apeiba membranácea.

Dendrochronology has been a tool of great importance when it comes to ecological studies and has allowed the study of climate and forests around the world. However, this technique was originally developed in temperate zones, which resulted in the bias that rings only occur in zones with seasons. For this reason, studies of growth rings in the tropics are minimal compared to studies outside of it. Due to the difference in the anatomical characteristics of the species within the tropics, it is necessary to create tools focused on these species that allow a greater development of dendrochronology in the tropics. Thus, the development of a method that allows segmenting the growth rings in the Apeiba membranacea species from computer vision techniques is proposed. The process begins by analyzing how to develop the acquisition of the images, finding that the best option for this is by scanning the images at a resolution of 1200 PPP, then the color spaces of these images were evaluated by experts criteria finding that the channels based on intensity are those that best reflect the anatomical characteristics, especially the RGB space, which presents with different levels the anatomical characteristics in each of its channels. Subsequently, different segmentation techniques were analyzed and it was found that the most appropriate is the use of a Ternausnet architecture in different batches weighing the final result. When validating the results against hand segmented images, a Jaccard index of 0.75, an accuracy of 0.85, a sensitivity of 0.85 and a specificity of 0.88 were obtained, concluding that the most appropriate way to address this problem is through the use of different models, trained based on daca one of the anatomical characteristics of the species.

• Autores: Sánchez Aguiar, Andrés Felipe
• Tipo de recurso: Work document
• Fecha de publicación: 2019
• Institución: Universidad Nacional de Colombia
• Repositorio: Universidad Nacional de Colombia
• Idioma: spa
• OAI Identifier: oai:repositorio.unal.edu.co:unal/75549
• Acceso en línea: https://repositorio.unal.edu.co/handle/unal/75549
• Palabra clave: Ingeniería y operaciones afines; Dendrochronology, Computer vision, Deep learning, Apeiba membranacea, Segmentation.; Dendrocronología, Visión por computador, Apeiba membranácea, Aprendizaje profundo, Segmentación
• Rights: openAccess
• License: Atribución-SinDerivadas 4.0 Internacional