The increasing demand on synchronized measurements obtained from Phasor Measurement Units (PMUs) has further increased potential threats, including data spoofing attacks. Such attacks on the synchronized measurements might greatly undermine the dynamic state estimation and important power system applications. In this work, an integrated framework consisting of Cyber Physical (CPS) testbed and multidimensional Convolutional Neural Network (CNN) simulation framework is proposed to classify the data spoofing attacks on the generated real world synchrophasor data. This work uses a source authentication-based detection technique by extracting the spatial fingerprint information from PMU data. This information is provided as inputs to the Multi-Dimensional CNN (MD CNN) simulation framework. Finally, a robust correction algorithm using Generative Adversarial Network (GAN) for reconstruction of spoofed PMU signal is proposed to eliminate bad measurements and enhance the integrity and reliability of synchrophasor data. Finally, the results of the proposed MD CNN classification framework and GAN based signal reconstruction algorithm are validated against the results obtained using traditional methods to validate the effectiveness of the proposed approach. Thus, the proposed framework aims at ensuring the cyber resilience of power grids to ensure its safe and reliable operation.