Metodologia para análise da deformação rígida e elástica entre diferentes bandas de uma imagem multiespectral

Abstract

As part of human evolution, among many other processes, technology leads to improve farming techniques through Unmanned Aerial Vehicles (UAV), focusing on better plan-tations and, consequently, on foods’ quality. This vehicles works on registering tillages’ photos by multispectral cameras. This paper proposes a methodology based on turningfunctions, to analyze the deformation of multispectral images obtained by UAV, classify-ing it through polygonal shapes comparison. One of the main challenges in this context is the classification and alignment of the deformations among the different bands, whereas combining many bands of multispectral images must provide better and diversified re-sults. Using the Control Point Selection (cpselect) tool, common points between pairs of different bands were manually obtained. Then, an algorithm was executed to obtain the best polygon formed by this points. Finally, using those best polygons, the method of turn-ing functions was applied for polygonal shapes comparison and then the dissimilarities were registered. Preliminary results has shown that both blue and green bands obtained better shape matching, whereas red and NIR were less similar (according to the adopted parameters). The purposed method was validated and performed good results too, as shown by some other authors. It was possible to verify which pair of bands worst or best matches one with another. However, the method couldn’t be concluded for another plantation types or another multispectral camera, once this method was validated just for a image’s series registered just by one camera. Therefore, future works with different types of plantations are necessary for a more generalized validation.