Hi @Vivek Kumar , Thanks for using Microsoft Q&A Platform.
I believe this can be tried using Custom Neural model. Custom neural models are capable of generalizing across multiple formats of a single document type. I suggest you create a single model for all variations of a document type using custom neural model. To the training dataset, add at least five labeled samples for each of the different variations.
Futher, consider a composed model, which is created by taking a collection of custom models and assigning them to a single model that includes all of your form types. When a document is submitted to a composed model, a classification step is performed to determine which custom model best represents the form presented for analysis. Compose model can be used for scenarios when you have documents of different types being submitted for analysis.
I hope this helps.
Regards,
Vasavi