How to improve accuracy of OCR?

Hello,

is there any way to label and train data in the custom model without manually updating ocr.json and label.json .

I'm currently facing issue with these characters,

Character having high failure frequency: -

1 (as I, J and T),

5 (as S),

D (as 0 and O)

Character having lower failure frequency: -

F (as E),

L (as 1),

Q (as O),

T (as 1 and 7),

Z (as 2)

is there any way to fix atleast character having high failure frequency?

Thanks