Deep Learning-Based Object Recognition and Design Information Extraction Method from Piping Isometric Drawings

Piping isometric drawings are critical deliverables in the plant industry, used for pipe construction and fabrication. However, manual review of these drawings is time-consuming and prone to human errors. We propose a method for detecting symbols and text in piping isometric drawings using a deep learning-based object detection model, and subsequently digitizing the extracted design information. The proposed method integrates the YOLOv10 model for simultaneous detection of both symbols and text, and utilizes the Tesseract OCR engine for text recognition. The extracted design information is then automatically saved in an Excel file. Experimental results show that the YOLOv10 model achieves high object detection performance, with a precision of 95.1% and a recall of 95.6%. These results confirm that deep learning techniques can be effectively employed to recognize symbols in piping isometric drawings and digitize design data. The proposed method has the potential to automate the review process, reducing human errors and improving design efficiency, thereby contributing to enhanced design quality in piping isometric drawings.