| I&T Solution |
Intelligent Lift and Escalator Reportable Incidents Documentation and Analysis System
(REF: S-0772) |
| Trial Project |
|
| Solution Feature |
- AI model building and training for OCR, image tuning, semantic classification, fine tuning performances with AI NLP analysis
- AI aided data extraction and user-friendly UI display for manual cross-checking and correction
- Comprehensive excel files for revised result presentation and on-going incident tracking
- Dashboard and summary report for data analysis and visualisation on incident trend description and correlation analysis
- Generate API calls to LEOS for form information storage
|
| Trial Application and Expected Outcome |
- Based on the exhaustive testing and fine-tuning of various OCR/Cloud-AI methods, we found that the content in printed format could be recognized perfectly. However, the best performance of hand-written content without optimization was only around 30% (in terms of edit distance).
- As the contents for most table columns were presented within a limited data space, intelligent correction and customized fine-tuning can be conducted for significant enhancement on performance and accuracy for hand-written content to the range between 60-85%, by training a predictive AI model for each column in the form.
- To provide an interactive data labelling function in the dashboard with an auto-selected range of raw images for users’ review and correction on the result. With the above adjustment and enhancement processes, it is highly possible that the accuracy of the system can ultimately reach 99.9%+.
- Customized enhancement of OCR for forms LE27/LE29: Edge detection/Tiles cropping/Pixel enhancement/Document distortion adjusting/Handwritten character splitting/OCR model testing and validation /Customized predictive AI correction
- Machine Learning/UI: Intelligent correction based on data dictionary and background knowledge/Trending analysis/Attribution Analysis/Clustering analysis of incidents data/Interactive data exploration and analytics/Interactive data labelling /Visualization of entity recognition/Basic map visualization on open platform/Multilingual processing for OCR/NLP analysis
|
| Additional Solution Information |
Mulit-OCR testing and evaluation report.pdf
|
| Info on I&T Solution Provider |
| Solution Provider | : | Bitex Limited | | Address | : | Unit 219, 2/F, Buidling 12W, No. 12 Science Park West Avenue, | | Contact Person | : | Susan Chen |
| Position | : | Vice President | | Tel | : | 90152162 | | Email | : |
susan@bithkex.com.hk |
|
