Journal bearing rotor systems are widely used in various industrial applications. Due to uncertainties, the rotor system requires efficient and accurate diagnosis system. To develop such diagnosis system, deep learning approach is used since deep learning techniques extract relevant health features automatically. In this study, we have revised the generation process for vibration images, which produces more images and require less producing time. The enhanced images are then used as the inputs to the convolutional neural network. To validate the enhanced vibration image, vibration signals from a journal bearing rotor testbed are used. The result shows that the efficiency has been improved while maintaining the high prediction accuracy for diagnosing health states.