CAD Matching Toolkit

This repository provides a comprehensive toolkit for CAD pattern detection and text matching in engineering documents.

Features

YOLO-Based Pattern Detection

• Detects engineering components in CAD drawings and P&IDs using YOLOv11
• Optimized for detecting multiple component types simultaneously
• Colorful annotations with unique colors for each component type
• High-resolution PDF processing with automatic tiling for large documents
• Supports concurrent processing on multi-core systems
• Generates detailed CSV reports of detection counts

Text Recognition

• OCR processing for text extraction from PDF documents
• String matching capability for finding specific text in engineering documents
• Highlights text occurrences with distinct colors for easy visualization
• Processes high-resolution images in overlapping patches for optimal text detection
• Generates CSV reports of text occurrence counts

Installation

Clone the repository and install the required dependencies:

git clone https://github.com/username/CAD_matching.git
cd CAD_matching
pip install -r requirements.txt

1. YOLO-Based Pattern Detection

Processing Pipeline

The pattern detection process works in two stages:

1. Preprocessing:

• Converts PDFs to high-resolution PNG images
• Applies binary filtering to optimize drawings for detection
• Handles large images via tiling to maintain full resolution

2. YOLO Detection:

• Segments large images for optimal detection
• Applies trained YOLO model to detect components
• Generates colorful annotations with component labels
• Creates CSV report with component counts

Usage

1. Pattern Matching

cd CAD_matching
python Pattern_match/main.py --input-cad Pattern_match/input_CAD --processed-dir Pattern_match/input_image --output Pattern_match/output_CAD

Command-Line Options

Option	Description	Default
--input-cad	Directory containing CAD drawings (PDF/images)	Pattern_match/input_CAD
--processed-dir	Directory for preprocessed images	Pattern_match/input_image
--output	Output directory for annotated images	Pattern_match/output_CAD
--model	Path to YOLO model weights	runs/detect/train/weights/best.pt
--dpi	Resolution for PDF conversion (dots per inch)	400
--workers	Number of worker threads for parallel processing	4
--size	YOLO detection segment size	1024
--overlap	Overlap between segments (0-1)	0.15
--conf	Confidence threshold (0-1)	0.25
--iou	IoU threshold (0-1)	0.45
--report	Path for CSV report	Pattern_match/detection_report.csv
--skip-preprocess	Skip preprocessing step	False

---

2. Text Recognition and Matching

The text recognition process consists of two main stages:

1. Preprocessing:

• Converts PDFs to high-resolution images
• Enhances text visibility with advanced filtering techniques
• Removes noise and optimizes contrast for better OCR results
• Text Detection and Matching:

2. Splits large images into overlapping patches for detailed text detection

3. Uses PaddleOCR to detect text in each patch

• Matches detected text against target words from a specified list
• Highlights and annotates matched words in the output images
• Generates detailed CSV reports with occurrence counts

cd CAD_matching
python String_match/main_text_det.py

Output

• Visual output with highlighted patterns in output directory
• CSV report (valve_reports.csv) showing pattern counts for each file
• Text report summarizing detection statistics

Command-Line Options

Option	Description	Default
--input-dir	Directory containing input PDFs	input_pdf
--filtered-dir	Directory for filtered/preprocessed images	filtered_img
--output-dir	Directory for highlighted output images	output
--target-file	Path to file containing target words	target.txt
--dpi	Resolution for PDF to image conversion	400
--patch-size	Size of image patches for OCR	512
--overlap	Percentage of overlap between patches	30
--case-sensitive	Use case-sensitive matching	False
--preprocess-only	Only run preprocessing step	False
--ocr-only	Only run OCR detection step	False