Diseño de un algoritmo para determinar el área de recursos hídricos empleando procesamiento digital de imágenes y redes neuronales.

Abstract

The purpose of this work was the segmentation and calculation of areas of water resources using Artificial Neural Networks (ANN) and Digital Image Processing (DIP). The use of Artificial Intelligence (AI) was used, which is the science that encompasses the branches used to achieve the objective set out in this work. A Convolutional Neural Network (CNN) was used, which had to be previously trained with satellite images of water resources, obtained from the Sentinel Hub web site and its EO Browser tool, these were pre-processed to later be labeled. Once the network was trained, a post-processing was applied to the images through Morphological Operations and Active Contours. Finally, the calculation of the area of the resource was carried out by counting pixels of the final segmentation obtained, the entire calculation process was implemented in a Graphical User Interface (GUI) in Matlab software. The results obtained by this software demonstrated a high precision of segmentation, exceeding a precision of 95%; the software was able to process satellite images taken in different spectra maintaining its high precision. It can do the calculations autonomously, therefore, it enables massive data processing, it is recommended to work in the Short-Wave Infrared (SWIR) spectrum for best results.