Geometrical correction of CBERS-4/PAN images with generalized models using as reference national system of land management data

• Autor(a) principal: Akiyama, Thales Shoiti
• Data de Publicação: 2018
• Outros Autores: Junior, José Marcato [UNESP], Tommaselli, Antonio Maria Garcia [UNESP]
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://dx.doi.org/10.11137/2018_2_358_368; http://hdl.handle.net/11449/232870
• topic: Cartography; Environmental preservation; Remote sensing

The orbital images have been widely used in several applications in the Earth observation context, which require different levels of detail and positional accuracy. The China-Brazil Earth Resources Satellite Program (CBERS) program was originated from a partnership between Brazil and China in the technical-scientific spatial sector. The CBERS-4 satellite is the fifth satellite of the CBERS Program and contains the PAN sensor, which collects panchromatic images with spatial resolution element (GSD-Ground Sample Distance) of 5 m. The researches related to the analysis of positional reliability and geometric correction of CBERS-4 images are still limited. Previous studies with CBERS-4 PAN images with different levels of processing indicate significant positional displacements of the georeferenced images, which are available by INPE (National Institute of Space Research). The positional displacements are incompatible with its GSD. The objective of this work was to investigate the application of generalized mathematical models in the geometric correction of CBERS-4 PAN images using rural properties limits of INCRA (Instituto Nacional de Colonização e Reforma Agrária) as control points. These limits are available for properties all over Brazil, which makes it possible to replicate the work at the national level. Images with different levels of previous correction (levels 1 and 2) were considered. Level 1 images are derived only from the application of radiometric calibration procedures, while level 2 images are level 1 images geometrically corrected from satellite orbital data information. In the experiments were considered 3 (three) images at level 1 and 1 (one) image at level 2. The following generalized models were adopted: Polynomials of order 1, 2 and 3; Projective and; Thin-plate spline (TPS). Generalized models have the advantage of not requiring knowledge of the system acquisition parameters, such as focal length, sensor size, among others. However, the generalized models require a significant amount of control points with uniform distribution throughout the image. For the geometric correction process were used different configurations of control points (30, 25, 20, 15 and 10) coinciding with the georeferenced rural properties in the Mato Grosso do Sul state, which presents accuracy higher than 50 cm. The geometric correction validation was performed from the RMSE (Root Mean Square Error) at checkpoints. The polynomial transformation of order 1 presented high values (higher than 10 GSD-50 meters) of RSME when compared to the other mathematical models, even considering 30 control points. The polynomial model of order 2 presented consistent behavior higher than the other models. Even when considering only 10 GCP presented RMSE between 1 and 2 GSD. In this model there is no significant improvement in the results, even increasing the number of control points. In the other models (TPS, Projection and Polynomial of order 3), there was a significant increase in RMSE when the number of points was reduced. The images used in this work cover part of the Mato Grosso do Sul state, which encompasses the most part of the Pantanal, considered a natural patrimony of humanity. Therefore, these orbital images contribute to the mapping and monitoring of their natural resources and, consequently, the protection of this patrimony.