In this paper, we propose a robust and reversible watermarking method for 3D mesh models based on Auto Diffusion Function (ADF) salient point extraction and strip-based partitioning. The method first utilizes ADF to extract significant geometric feature points of the model as anchor points for watermark embedding. A one-dimensional strip-based algorithm is extended to the three-dimensional space to achieve distributed and redundant embedding of watermark information into the multi-ring neighborhoods around multiple salient points. Experimental results show that the proposed algorithm exhibits strong robustness against geometric transformations such as vertex reordering, rotation, and translation. Regarding cropping attacks, the method maintains a high correlation coefficient for watermark extraction even under 20% cropping on complex models with high vertex density. More importantly, we introduces the concept of Practical Reversibility, achieving reversible operations by discarding part of the least significant bits. Experimental evidence shows that the proposed method can control both the RMSE and AVD between the recovered and original models at the order of \(10^{-12}\) , which is significantly below the defined Practical Reversibility threshold, thus achieving high-precision recovery.