In today's global manufacturing industry transitioning toward low-carbon and high-efficiency development, the recycling and remanufacturing of high-value equipment has become increasingly critical. As bolts are the most commonly used fasteners in the majority of equipment, research into autonomous bolt recognition and disassembly technologies plays a key role in reducing labor costs during the remanufacturing process. Initially, a robotic disassembly system and a circular fastener equipped with an upper guard plate were designed. Subsequently, given the difficulty of acquiring large-scale bolt datasets, a comparative analysis was conducted by training YOLO-based recognition models using both real-world datasets derived from the circular fastener and synthetic datasets. The training results demonstrate that properly processed synthetic datasets can be utilized to train models with satisfactory recognition capabilities, effectively reducing the reliance on collecting extensive real-world data. Furthermore, based on the bolt recognition algorithm, disassembly target localization and autonomous disassembly planning were implemented, followed by physical experiments in real-world environments. The experimental results validate the high success rate and accuracy of the proposed recognition and planning framework.