Modelos estadísticos para la detección de patrones en medio ambiente y economía.

Abstract

Oil spills cause serious damage to nature. The literature reviewed argues that early detection and monitoring of spills is the most important management. It is important to know precisely the exact time and place of spills. In addition, knowing what type of oil is spilled can help decision makers choose an appropriate method for cleanup. Remote sensing is considered one of the most effective methods for detecting oil spills. SAR remote sensing is based on the interaction of electromagnetic waves from radar with matter waves from the sea surface. Active remote sensing sensors overcome the limitations of passive sensors in the face of weather and time conditions. This is the reason for selecting the satellite images of the missions captured by Sentinel-1 and available in the repositories of the European Space Agency (ESA). To extract these images, a database of oil spill incident records with their location and exact date of occurrence was compiled. On the other hand, a method for the preprocessing of satellite images is proposed with special attention to the treatment of speckle noise. Then, the cooccurrence matrix and second order textures are calculated using SNAP linked to Python. From the extracted features a Principal Component Analysis is performed to reduce the dimension of the feature set and finally the unsupervised learning algorithm $k-$means is applied to cluster pixels.