Clasificación de señales caóticas sintetizadas en imágenes por medio de técnicas de aprendizaje automático científico

Abstract

In this work we propose a method of image synthesis from signals, based on the phase space reconstruction method proposed by Takens, which fits with the objective of Scientific machine learning by integrating a neural network for the detection of chaos in signals, so that it can facilitate the identification of unknown dynamic systems. Understanding that chaos can be detected in phase space, we synthesize images that have information about the reconstruction of this space. A chaotic signal classifier model is tuned with 94.3% precision and 94.1% accuracy, trained with the transfer learning method. This model supports the proposal of integrating machine learning algorithms in questions from the area of science, associating these two areas of knowledge.