Métodos para estimativa de imagens NIR a partir de imagens de câmeras RGB

• Autor(a) principal: Lima, Daniel Caio de
• Data de Publicação: 2020
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Repositório Institucional da UFSCAR
• Texto Completo: https://repositorio.ufscar.br/handle/ufscar/12450
• topic: Sensoriamento Remoto; Remote Sensing; Estimativa de Imagem NIR; NIR Image Estimative; Deep Learning; Pix2Pix; KNN; Assinatura Espectral; Spectral Signature; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO

Precision Agriculture involves the use of technology for management and decision making to allow rural producers to get better production results. To gain the desired effect, it is necessary a huge data collection and processing of the cultivated areas, being Remote Sensing, using special sensors attached to drones, a powerful alternative for this task. Vegetation Indices obtained with arithmetic equations, are used to highlight vegetation cover density variations using near infrared (NIR) images and visible spectrum images, being useful to assess biomass and productivity estimation from a crop, for example. The main problem is the elevated costs of drones and principally of sensors that acquire NIR images, making the use of this technology, by small producers, unattractive or difficult. The objective of this work is to propose a method to estimate NIR images using ordinary photographic camera RGB images that could eliminate the use of specific sensors and the need to make alterations in commom cameras, thus making this technology cost lower. To achieve this goal, we propose the use of a Deep Learning architecture (Pix2Pix) and a spectral signature based, KNN classification and a weighted sum by proximity degree of k nearest reference signatures method, producing a new spectral signature. These methods are described and evaluated in this document. Results showed that the two methods investigated can be used to estimate NIR images, showing high similarity to real NIR images.