Protection of high voltage transmission lines is one of the crucial problems in the power system engineering. Accurate and timely detection and identification of transmission line short circuit faults can considerably improve and simplify their recovery process and hence save the costs associated with the downtime of a power system. Hence, it is essential that a robust and reliable fault diagnosis system completes its operation within an acceptable time window after fault occurrence in the presence of uncertainties and disturbances in the system. The significant costs of mistakenly detected or undetected faults based on the conventional techniques motivate us to present a robust detection and identification system by using the convolutional neural networks. The robustness of this technique is analyzed for the variations of the phase difference between two connected buses, fault resistance, source inductance fluctuations, fault inception angle, local bus voltage fluctuations, and measurement noises. The time delay analysis is also conducted to indicate that the presented technique is able to detect, identify, and estimate the location of faults before tripping relays and circuit breakers disconnect a faulty region.