Implementación de una Metodología basada en una Red de Aprendizaje Profundo para Segmentación Semántica en Imágenes de Percepción Remota
In the recent years, vegetation has been changing due to natural phenomena and human activity. Considering these facts, studies were carried out in order to detect changes in the vegetation species and their environment. Remote sensing is a technique for capturing images of areas and objects without...
| Autor: | |
|---|---|
| Formato: | tesis de maestría |
| Estado: | Versión aceptada para publicación |
| Fecha de publicación: | 2022 |
| País: | México |
| Recursos: | Instituto Nacional de Astrofísica, Óptica y Electrónica |
| Repositorio: | Repositorio Institucional del INAOE |
| Idioma: | español |
| OAI Identifier: | oai:inaoe.repositorioinstitucional.mx:1009/2413 |
| Acesso em linha: | http://inaoe.repositorioinstitucional.mx/jspui/handle/1009/2413 |
| Access Level: | acceso abierto |
| Palavra-chave: | info:eu-repo/classification/Aprendizaje profundo/Deep learning info:eu-repo/classification/Segmentación semántica/Semantic segmentation info:eu-repo/classification/Percepción/Perception info:eu-repo/classification/cti/1 info:eu-repo/classification/cti/25 info:eu-repo/classification/cti/2512 |
| Resumo: | In the recent years, vegetation has been changing due to natural phenomena and human activity. Considering these facts, studies were carried out in order to detect changes in the vegetation species and their environment. Remote sensing is a technique for capturing images of areas and objects without direct contact with them and their environment. These aerial inspections of vegetation regions can retrieve information for immediate decision-making to protect endangered species and their environment. Furthermore, the use of unmanned aerial vehicles (UAV) helps to obtain images quickly, allowing for analysis using artificial intelligence (AI) to automatically collect data and extract patterns. In this way, accurate information extraction is possible and can be used for better species control and protection. Also, semantic segmentation is a task developed with AI and the application of deep learning algorithms for complete scene comprehension. Many issues of semantic segmentation are resolved with deep learning architectures based on convolutional neural networks (CNN). This thesis proposes the application of a deep learning network for semantic segmentation of sensing remote images. The goal of this research is the automatic identification of vegetation species. In this work, the automatic semantic segmentation process started with the labeling task of images belonging to a specific vegetation data set: palm, pine, and orange trees. Then, this information is used for the training and validation process. Later, a deep learning technique was applied, and a model based on the pre-trained VGG16 architecture was used to automatically carry out the semantic segmentation on images. Finally, the specific classes were semantically segmented with an accuracy prediction ranging from 91 % to 96.5 %. These results have better precision than similar approaches reported in the literature. |
|---|