SurveyNet: A Unified Deep Learning Framework for OCR and OMR-Based Survey Digitization.

Area of Science:

• Computer Science
• Data Science
• Artificial Intelligence

Background:

• Manual data entry from survey sheets is a significant bottleneck in research and public policy.
• Existing Optical Character Recognition (OCR) and Optical Mark Recognition (OMR) systems are often separate and struggle with real-world survey data variability.

Purpose of the Study:

• To develop a unified deep learning framework, SurveyNet, for automated digitization of complex survey responses.
• To integrate Optical Character Recognition (OCR) and Optical Mark Recognition (OMR) capabilities into a single model.

Main Methods:

• SurveyNet processes handwritten digits and various mark types (ticks, circles, crosses) within a single deep learning model.
• A novel dataset, SurveySet, comprising 135 real-world survey forms, was created for training and evaluation.

Main Results:

• SurveyNet achieved classification accuracy ranging from 50% to 97% across different tasks.
• The model demonstrated strong performance even on small and imbalanced datasets, indicating robustness.

Conclusions:

• SurveyNet provides a scalable solution for automating survey digitization, reducing manual errors, and enabling faster data analysis.
• The framework has broad applicability in fields like consumer research, public health, and education.