Bibliographic citations
Yupanqui, O., (2024). Generación de un modelo para la detección de extracción selectiva de árboles en el departamento de Ucayali mediante aprendizaje automático [Universidad Nacional Agraria La Molina]. https://hdl.handle.net/20.500.12996/6891
Yupanqui, O., Generación de un modelo para la detección de extracción selectiva de árboles en el departamento de Ucayali mediante aprendizaje automático []. PE: Universidad Nacional Agraria La Molina; 2024. https://hdl.handle.net/20.500.12996/6891
@misc{renati/1116286,
title = "Generación de un modelo para la detección de extracción selectiva de árboles en el departamento de Ucayali mediante aprendizaje automático",
author = "Yupanqui Carrasco, Osmar",
publisher = "Universidad Nacional Agraria La Molina",
year = "2024"
}
Currently, there is no national research regarding satellite detection of selective tree extraction. This is due to the difficulty of obtaining geospatial and cartographic information related to this activity and the short period of time in which it is possible to detect it. The present study seeks to generate a model based on machine learning algorithms, capable of detecting selective tree extraction in the region of Ucayali. The model consists of the application of a metamodel, created from satellite data of Sentinel-2 spectral indices from 2019, whose predictions are used to generate density bands, which function as input for the final model, together with Sentinel-2 spectral indices from 2020, and an image of distances to active roads/camps. This model was generated in a Google Colab virtual machine, where the training was carried out, while its application was carried out in the cloud (Google Earth Engine). In both models, the data set was divided into training and testing, with a proportion of 70% and 30% respectively. Eleven different types of algorithms were tested based on a “k-fold” cross-validation with 10 repetitions. The Extra Trees Classifier was selected, as it had the highest percentage of success (greater than 98%). The hyperparameters of the algorithm were optimized and the model was evaluated based on the test data set. Finally, the model was validated with a stratified sampling design, with a random distribution of points. An overall success rate of 75.42% was obtained. The final model has the potential to be used in other regions besides Ucayali with fast results due to cloud computing.
This item is licensed under a Creative Commons License