Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Billet Number Recognition Based on Test-Time Adaptation

Yuan Wei, Xiuzhuang Zhou

TL;DR

The paper tackles real-time billet number recognition under production conditions, where distribution shifts and damaged characters hinder accuracy. It introduces a two-pronged approach: (i) test-time adaptation by minimizing entropy $H(oldsymbol{ ilde{y}})$ during testing to adapt a DB+SVTR pipeline without labels, updating only BatchNorm parameters, and (ii) inference-time priors based on billet-number encoding rules and a damaged-character correction for CTC to enforce valid outputs and recover from missing characters. The method uses Differentiable Binarization (DB) for detection and SVTR for recognition, with post-processing that enforces encoding constraints and improves CTC handling of blanks in damaged regions. Experiments on machine-printed and handwritten billets show substantial gains, with accuracy improving from baseline 0.58 to 0.79 on machine-printed and from 0.23 to 0.70 on handwritten numbers when both TTA and prior-knowledge constraints are employed, demonstrating practical impact in industrial settings.

Abstract

During the steel billet production process, it is essential to recognize machine-printed or manually written billet numbers on moving billets in real-time. To address the issue of low recognition accuracy for existing scene text recognition methods, caused by factors such as image distortions and distribution differences between training and test data, we propose a billet number recognition method that integrates test-time adaptation with prior knowledge. First, we introduce a test-time adaptation method into a model that uses the DB network for text detection and the SVTR network for text recognition. By minimizing the model's entropy during the testing phase, the model can adapt to the distribution of test data without the need for supervised fine-tuning. Second, we leverage the billet number encoding rules as prior knowledge to assess the validity of each recognition result. Invalid results, which do not comply with the encoding rules, are replaced. Finally, we introduce a validation mechanism into the CTC algorithm using prior knowledge to address its limitations in recognizing damaged characters. Experimental results on real datasets, including both machine-printed billet numbers and handwritten billet numbers, show significant improvements in evaluation metrics, validating the effectiveness of the proposed method.

Billet Number Recognition Based on Test-Time Adaptation

TL;DR

The paper tackles real-time billet number recognition under production conditions, where distribution shifts and damaged characters hinder accuracy. It introduces a two-pronged approach: (i) test-time adaptation by minimizing entropy during testing to adapt a DB+SVTR pipeline without labels, updating only BatchNorm parameters, and (ii) inference-time priors based on billet-number encoding rules and a damaged-character correction for CTC to enforce valid outputs and recover from missing characters. The method uses Differentiable Binarization (DB) for detection and SVTR for recognition, with post-processing that enforces encoding constraints and improves CTC handling of blanks in damaged regions. Experiments on machine-printed and handwritten billets show substantial gains, with accuracy improving from baseline 0.58 to 0.79 on machine-printed and from 0.23 to 0.70 on handwritten numbers when both TTA and prior-knowledge constraints are employed, demonstrating practical impact in industrial settings.

Abstract

During the steel billet production process, it is essential to recognize machine-printed or manually written billet numbers on moving billets in real-time. To address the issue of low recognition accuracy for existing scene text recognition methods, caused by factors such as image distortions and distribution differences between training and test data, we propose a billet number recognition method that integrates test-time adaptation with prior knowledge. First, we introduce a test-time adaptation method into a model that uses the DB network for text detection and the SVTR network for text recognition. By minimizing the model's entropy during the testing phase, the model can adapt to the distribution of test data without the need for supervised fine-tuning. Second, we leverage the billet number encoding rules as prior knowledge to assess the validity of each recognition result. Invalid results, which do not comply with the encoding rules, are replaced. Finally, we introduce a validation mechanism into the CTC algorithm using prior knowledge to address its limitations in recognizing damaged characters. Experimental results on real datasets, including both machine-printed billet numbers and handwritten billet numbers, show significant improvements in evaluation metrics, validating the effectiveness of the proposed method.

Paper Structure

This paper contains 15 sections, 1 equation, 5 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Billet Number Recognition
  4. Test-Time Adaptation
  5. Method
  6. Text Detection and Recognition Networks
  7. Test-time Adaptation Method
  8. Inference with Prior Knowledge Constraints
  9. Automatic Correction of Recognition Results Based on Encoding Rules
  10. Correction of CTC Algorithm Under Character Damage
  11. Experiment
  12. Dataset
  13. Experiment Settings
  14. Experiment Results
  15. Conclusion

Figures (5)

  • Figure 1: Overview of our method.
  • Figure 2: SVTR Network Architecture.
  • Figure 3: The relationship between entropy and error rate.
  • Figure 4: Architecture of the CTC algorithm.
  • Figure 5: Misrecognition of the CTC algorithm in the case of damaged characters. The left image shows the case where the recognition box is aligned with the target character, but the character is damaged. The right image shows the case where the recognition box is not aligned with the target character. It can be seen that both cases yield the same recognition result.