With more sensors being installed by utilities for accurate control of power grids, there is a growing risk of vulnerability to sophisticated data integrity attacks such as false data injection (FDI), circumventing current bad data detection schemes resulting in inaccurate state estimation solutions. While diverse automated detectors to battle FDI have been grown, such methodologies underestimate the strong analytical abilities of humans. This is while most proposed automated methods need observant human control. Although automated methods provide opportunities to improve scalability, humans can cope with exceptions and new attack trends. In this paper, to address the ever-increasing cyber-attack challenge in power systems, a visualization based attack detection framework using deep learning techniques is developed to provide human security researchers with improved techniques to uncover trends, identify outliers, recognize correlations, and communicate their results. Specifically, we first encode multivariate systems state time-series data into 2D colored images and then utilize a carefully designed deep convolutional neural network (CNN) classifier. The proposed method is developed to allow network operators to immediately capture and visually understand the statistical features of a network attack at a glance. The proposed method has been evaluated on the IEEE 14-bus and IEEE 118-bus systems. Our experiments show that the proposed framework accomplishes high classification accuracy.