Comparative analysis of classification methods of high -resolution images for the generation of thematic geo -information at detailed scale

Currently, a large part of the thematic geo -information (coverage, vegetation indices, topography, etc.) is still generated from satellite images or aerial photographs of several meters of space resolution, this prevents obtaining geoinformation at a detailed scale. In various projects focused on research, territorial planning and other themes, detailed geoinformation is vital for decision -making based on a representation closer to reality. Additionally, detailed geo -information gives the possibility of venturing applications that require a lot of precision such as territorial planning in detail, plant health, quantification of plant species, etc. On the other hand, different platforms provided with sensors, are increasingly, capturing multispectral images with high temporal resolution, and high spatial resolution in the order of the centimeters. However, the processing of this large number of images brings new challenges due mainly to the large number of pixels obtained in multiple layers. In this context it is important to compare different methods: traditional and new trends, and evaluate the results of the image classification process for thematic cartography purposes; Considering in processing, possible new requirements in high performance computing. The objective of this research is to compare different methods of classification of high -resolution images, for the generation of thematic geo -informed at a detailed scale and evaluate the results in terms of precision and computational efficiency, to provide more precise thematic geo -information to decision makers. The methods to be compared consider pixel -based algorithms, object -based image analysis (OBIA), and use of artificial intelligence techniques specifically Deep Learning algorithms. Methodologically we will work with a case study on the Ecuadorian interandin region for the generation of coverage geoinformation from multispectral orthoimagnes (RGB and nearby infrared), 10 cm pixel size. These images come from Aerophotogertric flight carried out by the Stereocarto company in 2012, currently rest in the Pomas-University's repository of Cuenca and are available for research purposes. The results obtained from each classification method will be validated contrasting: computational time; And its precision to the extent that they reflect reality, using pre-existing geo-information (coverage, level curves, etc. scale 1: 1000) of the same flight date, which has been validated with field work by domain expert technicians. This research will deepen the efficiency of the classification algorithms applied to high -resolution images of the Ecuadorian interandin region. In addition, different classification methods analyzed will be implemented as a software prototype (using Opensource tools) so that they can potentially be executed on other entry images and thus provide experts in the domain or decision makers of a tool that allows the best choice for the generation of thematic geo -information at a detailed scale. The main contribution to the state of the art constitutes the possibility of suggesting the most appropriate method to generate detailed thematic geinformation when working with high -space mountainous images of mountainous areas. Finally, the research project will disseminate the partial and/or final results in national and international congresses with dissemination in magazines with impact factor.

OUTSIDE THE CALL FOR PROPOSALS INTERNAL FUNDS