Product Manufacturing Information Recognition from Engineering Drawings

Product Manufacturing Information (PMI) plays a vital role in engineering drawings for intelligent manufacturing, as its precise and efficient extraction is essential for automating process planning. This research introduces a hybrid approach that merges computer vision techniques with deep learning to address the difficulties in extracting PMI, including dispersed text layout and intricate technical symbols. The proposed framework features a multi-phase processing pipeline, which involves object detection, rotation correction and text extraction. Notably, it utilizes oriented bounding box detection along with a four-category classification model to greatly enhance the accuracy of identifying PMI regions. Additionally, the study presents a combined text analysis structure that integrates object detection with Optical Character Recognition (OCR), enabling the effective extraction of specialized elements such as dimensional tolerances and surface texture annotations.