The integration of data at different frequencies, collected with different antennas or with the use of swept frequency radars opens up interesting perspectives in the study of the subsurface at different resolutions. In this context, Deep Learning (DL) can have a relevant role for data merging, in order to allow the use wider frequency data during processing and interpretation phases. The proposed methodology is a semi-supervised Deep Learning algorithm based on Bi-Directional Long Short Term Memory to automatically merge variable numbers of data sets at different frequency. A further constraint is given by the introduction of a user defined area in which we want to perform the merging procedure, allowing the interpreter to directly constrain the process. The training of the Neural Network is made directly on the inference data by minimizing a custom loss function based on the L2 norm of all the input data, weighted over the custom merging area, and the single output trace. The inference of the trained Neural Network is applied on the same data. The proposed algorithm is tested on synthetic data and on real data, showing very robust performances even in high noise conditions.
