This thesis investigates the possibility of using a Sia-mese Neural Network in order to create an algorithmfor global localization in the context of Mars rovers. Thethesis details the whole process of creating the NeuralNetwork: from the acquisition and processing of twodatasets to the creation of the model and the tuning ofits hyperparameters. Then, the model is tested and itsresource usage analyzed. The whole model only takesup to 9 megabytes of space and is able to give predic-tions in 18 milliseconds. The thesis also shows how themodel can be used in a global localization algorithm byimplementing a sliding window approach that can becompared with previous works. This sliding windowapproach shows some promising results and seems toperform better than the previous solution.