With the development of data element transformation, more and more organizations and individuals are starting to share data, and the risks faced by data sharing are also increasing day by day. Database watermarking as an important copyright protection and traceability technology, plays an important role in promoting data sharing. Existing database watermarking technologies usually rely on database primary keys or other tuples in the database, Insufficient universality. This article proposes a database watermarking scheme that does not rely on primary keys for attack resistance. In the watermark embedding stage, the watermark is first generated using the double finger method, then divided into multiple sub watermarks and inserted into the dataset according to certain rules. In the watermark detection stage, the check bit is checked first, and after passing the check, the sub watermarks are extracted, and then the embedded watermark is selected using a majority voting mechanism. Finally, By conducting simulation experiments on real datasets, it is verified that this scheme has good resilience against tuple deletion attacks, tuple addition attacks, tuple modification attacks, attribute deletion attacks, and Combination attack.