This work describes the percolation phenomena in different structures through deep neural networks and previously calculated statistical data of percolation. Despite being relatively simple and easy to calculate at small scales, the percolation process is computationally time-consuming at large scales; here, a significant computation is necessary to determine if a cluster percolates or not. We propose to train deep neural networks on small systems and scale to large systems. Our results show a reasonable accuracy rate on recognition of images, particularly on fully convolutional neural networks for the continuum case, a recent improvement on classical convolutional neural networks, improving the recognition of percolation phenomena, portability, and scalability.