Controller Area Network (CAN) is a ubiquitous broadcast protocol facilitating real-time communication among electronic control units (ECUs) in vehicles. Despite its importance, the proprietary nature of CAN message formats across different vehicle models and manufacturers poses significant challenges for signal interpretation and data analysis. This paper introduces AutoDBC, a novel end-to-end CAN-Bus Data Domain Intelligent Segmentation algorithm designed to address the limitations of current manual and automated CAN data reverse engineering methods. The AutoDBC framework innovatively extracts features at three levels-bit, byte, and message-to capture the nuanced characteristics of CAN data, accounting for the mixed endianness commonly found in automotive communications. Unlike existing approaches that rely solely on bit-level features such as bit flipping rates, AutoDBC considers the spatial-temporal relationships across bytes and messages, providing a more comprehensive understanding of the data. At the core of AutoDBC is a custom deep learning model that enhances the precision of signal boundary identification. Trained on a diverse dataset of real-world CAN bus data collected from various personal vehicles, this model demonstrates a remarkable improvement in accuracy compared to state-of-the-art (SOTA) methods. Our evaluation shows that AutoDBC achieves a 10–30% increase in precision for signal boundary and signal field identification, a significant leap forward in the field of automotive data analysis.