Bibliographic citations
Salazar, I., (2018). Evaluación y análisis de algoritmos para la estimación de áreas de síntomas visuales de infección en hojas de plantas usando imágenes digitales [Tesis, Universidad Nacional de Ingeniería]. http://hdl.handle.net/20.500.14076/17018
Salazar, I., Evaluación y análisis de algoritmos para la estimación de áreas de síntomas visuales de infección en hojas de plantas usando imágenes digitales [Tesis]. : Universidad Nacional de Ingeniería; 2018. http://hdl.handle.net/20.500.14076/17018
@misc{renati/709831,
title = "Evaluación y análisis de algoritmos para la estimación de áreas de síntomas visuales de infección en hojas de plantas usando imágenes digitales",
author = "Salazar Reque, Itamar Franco",
publisher = "Universidad Nacional de Ingeniería",
year = "2018"
}
Quantifying ill tissue areas in leaves of plants is an important procedure in agriculture due to its used in crops monitoring and decision making. However, this is a manual procedure which is time-consuming and highly consuming. Therefore, its automation is widely considered. In this work, we review some state-of-the-art methods performed to solve this problem. In addition, a new methodology is proposed. This new approach uses the Simple Linear Iterative Clustering (SLIC) algorithm, created in the Süsstrunk laboratory in 2010, to group nearby pixels with similarity of color in regions called superpixels. Then, color characteristics of a set of super pixels taken from representative samples were used to train artificial neural networks (ANNs) that were used as classifiers of these superpixels. The parameters of these networks were tuned in a heuristic way choosing the best classification network, which was used to obtain the automatic segmentation of the diseased areas. The performance of the classifier was measured by comparing automatic segmentations with those manually elaborated from a database with public and own images divided into nine groups according to the visual symptoms and the plant. This comparison was made with some techniques reported in the literature based on the error of the diseased area found and the F-score metric for the segmentations. The median of the area error percentage obtained was always less than 11% and the average F-score obtained was 0.67, which is higher than the averages of 0.57 and 0.58 of the other two approaches used here for the comparison.
This item is licensed under a Creative Commons License