<p>This paper addresses the challenge of remote driving capability identification in intelligent vehicles to enhance driving safety under communication delays and packet loss. A multimodal data-driven method based on graph transformation and hybrid convolutional neural network (Hybrid-CNN) is proposed. Specifically, a risk factor analysis is conducted using sensed driving data to establish an evaluation framework for remote driving takeover capabilities. To enable finer feature extraction, a graph transformation technique utilizing the Gramian angular field (GAF) is introduced, converting remote driving capability data into a two-dimensional (2D) representation. Furthermore, prediction errors are incorporated into a neural iterative updating mechanism to refine the model. Finally, a Hybrid-CNN is designed to concurrently extract global features from raw one-dimensional (1D) data and correlation features from the transformed 2D data. The simulation results show that the proposed method can effectively identify remote driving capabilities and improve the safety of the driving process.</p>