Implementación de las transformadas ridgilet, contourlet y curvelet para el análisis multitemporal, identificación y predicción del cambio de uso y cobertura del suelo en los municipios de Mosquera, Facatativá, Chía, El Rosal, Funza y Madrid-Cundinamarca usando imágenes landsat

Abstract

This document discusses the application of four algorithms, namely, contourlet, curvelet and ridgelet, wavelet fusion, which are intended to improve edges and spatial resolution respectively in an image, which were applied to images of the landsat 7 sensors and landsat 8 merged, in order to perform supervised and unsupervised classifications, in order to determine changes in land uses and coverage, with an emphasis on urban areas. The fusion was carried out through the application of several wavelet transforms, among them, main components, bior, rbio, haar, daubechies, with different levels of decomposition. In the case of the landsat 7 sensor, a gaps correction was made due to the error that this presents through the envi and erdas software, obtaining better results with the first one. The best results for the case of the landsat 7 sensor was bior 2.2 and for the landsat 8 rbio1.3, both in the decomposition level four. As for the application of contour improvement algorithms, namely, contourlet and ridgelet-curvelet, they were developed using the erdas software and the best algorithm for both sensors turned out to be the contourlet transform. The images fused with the wavelet transform (bior2.2 and rbio1.3) and the generated images (using the contourlet, curvelet and ridgelet transforms) were evaluated and analyzed quantitatively and qualitatively. The quantitative methods in the present analysis include ergas, rase, Universal Quality (Qu) and correlation coefficient (CC), whereas the qualitative method takes into account that spectral quality is not lost visually. The fusion of images and the implementation of the transformations were done with the MatLab® software, which provides the following Toolbox: Wavelet Toolbox, Image Processing Toolbox, Toolbox Contourlet and source code for Curvelet and Ridgelet. In the case of classifications supervised by vector support machines and decision trees, the best classification was the first. The highest Kappa coefficients were obtained from the merged image, followed by the merged Contourlet and finally the Contourlet. When carrying out the multitemporal analysis through the classifications made by vectorial support machines, some changes in land use and coverage were evidenced, especially the growth of the urban sector along the surrounding municipalities. A prediction was made for the year 2040 of the artificialized territories class that refers to the urban classes, and I throw a quite high value, which concludes that despite the dynamics of the urban land indicate an excessive growth of this, the value proposed by the prediction exceeds the limits of the territory, so it is necessary, if you want to go into depth in this topic, think about other variables and factors that affect this behavior, and that were not taken into account for this project. In conclusion, the mergers made show a remarkable change in the spatial resolution of the images, allowing to better identify areas, objects or coverage that maybe before the application of these were not so clear. Likewise, the classifications achieved their purpose allowing identifying different coverages through automatic tools or direct interaction with the user, allowing to identify the uses and coverage of the floor of the study area with an acceptable percentage of confidence taking into account the limitation that we have in terms of images that cover the study area.