This work investigates the use of Deep Learning techniques to detect galaxy mergers using astronomical imaging data from photometric surveys. We analyse three Convolutional Neural Networks architectures found in the literature and compare their performances training from scratch and using transfer learning. The models outperformed previous automatic detection methods and transfer learning showed a slightly better performance over training from scratch. A reliable approach to detect galaxy mergers is presented with different models achieving a precision of 0.97 on the dataset used.