Intelligent Recognition of Geometric Tolerance Symbol Annotation Blocks

In previous mechanical design drawings, the geometric tolerance symbol annotation block was directly annotated on the mechanical design drawing by engineers, resulting in the symbol annotation block being non-editable and non-shareable. To achieve digitization, this study proposes an improved Swin_YOLOv5_GE model based on YOLOv5 to realize the positioning of the symbol annotation block. Compared with the original YOLOv5, the accuracy is increased by 4%. Then, image processing is performed on the located and cropped symbol annotation block to achieve denoising and tilt correction of the symbol annotation block. Finally, it is input into CRNN for the recognition of the symbol annotation block. The final experimental results prove that the method in this paper can obtain an overall recognition accuracy of 94%, proving the feasibility of this method.