Scrape datatsheets with optical AI

[fl4p]

charts and test conditions

CSD19506KCS copy.pdf
sample

Qgd - V chart

Characteristics

• Dynamic Characteristics
• Diode Characteristics
• Thermal Information
  

[fl4p]

Infineon IQD016N08NM5 data sheet:

[fl4p]

google 'github extract values from charts'

• https://github.com/pratistha-gupta/Data-Extraction-From-Graphs PDF, OpenCV
• https://github.com/TheJaeLal/LineFormer
• https://github.com/MasterAI-EAM/GraphMaster

[fl4p]

onsemi NVMTS6D0N15MC

[fl4p]

https://toshiba.semicon-storage.com/eu/semiconductor/knowledge/faq/mosfet/electrical-characteristics-of-mosfetscharge-characteristic-qg-qg.html

note that there are 2 definitions for Qgs2:

• the remaining charge after the miller plateu (NVMTS6D0N15MC)
• the charge threshold until the plateau (FDP027N08B)
  

[fl4p]

datasheet TPH5R60APL lists Qgs1 wich is the charge until the start of miller plateau

[fl4p]

Update:

• ocrmypdf delivers good results
• wan to try pix2text, which generates markdown tables.
• tesseract with tables (TODO link)
• train tesseract or other custom models (pix2text)

Ocr Issues:

• most of the missing fields are because OCR struggles with reading symbols properly (Qg, Vsd ...) .
• tabular gets table wrong
• Symbols are no words, so no support from dictionary.
• tesseract supports user word list, this helps with Vplateau but not with Qgs, Qgd.
• Numerical OCR can be improved
• table structure sometimes generates characters