Intero-The Sniffers
Specializes in pipeline integrity, emission monitoring, and energy efficient solutions, helping industries ensure complience, safety and sustainability.
Balen, Belgium
1991
Energy, Oil & Gas
150-200 employees
Challenge
Enhancing deep learning models for the automated detection and recognition of symbols and text in Piping and Instrumentation Diagrams (P&IDs), optimizing accuracy and efficiency in industrial documentation processing.
Results
Improved the Deep Learning Models for Text Detection and Recognition.
Text Detection with Faster R-CNN ResNet-50: Optimized Faster R-CNN with a ResNet-50 backbone to improve text localization in complex P&ID diagrams, achieving better precision and recall.
Text Recognition with TransformerOCR: Fine-tuned Microsoft’s TransformerOCR for enhanced text recognition, improving accuracy on domain-specific P&ID symbols and text variations.
Optimizing the LEC placement
Kris Huybs
ICT Director at Intero - The Sniffers
Summary
After retraining of the text detection and text recognition models the overall Impact was:
TP (True Positives) significantly improved across all models, ranging from +18.47% to +162.74%.
FP (False Positives) saw a notable reduction in all models, with improvements between -5.11% and -45.13%.
FN (False Negatives) decreased dramatically in three cases, with the most significant reduction in the Brunei LNG model (-100%).
Mean Character Error Rate (CER) decreased by up to 46.23%, highlighting improvements in recognition accuracy.