Optical Reader Neural Network Inference for Code Restoration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optical information reading devices face challenges in reducing calculation load and size while maintaining reading accuracy, particularly when incorporating machine learning for code image restoration, which hinders the development of handheld devices and increases processing time in stationary devices.
Innovation Solution
The implementation of a neural network configured using pre-learned structures and parameters from defective and ideal images, allowing for inference processing without additional learning, and the option to apply an image restoration filter based on user settings to optimize processing speed and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning is incorporated into the optical information reading device to restore code images, then reading accuracy is improved, but the calculation load becomes extremely heavy and device size increases
Solution Approach 1:
The neural network model is trained in advance using a training device, and only the trained model (structure and parameters) is stored in the optical information reading device. During operation, the device performs inference processing using the pre-trained model rather than performing training itself, which dramatically reduces the calculation load and device complexity while maintaining high reading accuracy.
Solution Approach 2:
The patent introduces a division of labor between a training device (intermediary) and the optical information reading device. The training device performs the heavy computational work of model training, while the reading device uses the pre-trained model for lightweight inference. This intermediary approach allows the reading device to achieve high accuracy without bearing the full computational burden.
2Measurement precision
If machine learning is incorporated into the optical information reading device to restore code images, then reading accuracy is improved, but processing speed decreases due to heavy calculation load
Solution Approach 1:
The neural network model is trained in advance using a training device, and only the trained model (structure and parameters) is stored in the optical information reading device. During operation, the device performs inference processing using the pre-trained model rather than performing training itself, which dramatically reduces the calculation load and device complexity while maintaining high reading accuracy.
3Measurement precision
If the neural network processes the entire code image, then restoration quality is improved, but the calculation load increases
Solution Approach 1:
The code image is divided into multiple regions, and the neural network processes only the region containing the code rather than the entire image. This segmentation approach maintains restoration quality for the code area while significantly reducing the calculation load by excluding unnecessary processing of background and other non-code regions.
Solution Approach 2:
The patent applies different processing strategies to different regions of the image. The region containing the code undergoes neural network-based restoration with high computational resources, while other regions are processed with simpler methods or skipped entirely. This local quality approach optimizes the balance between restoration quality and calculation load.
Data Source
AI summary
To improve reading accuracy by enabling a read image to be restored by machine learning while reducing a calculation load in an optical information reading device. An optical information reading device includes: a storage part that stores a structure and a parameter of a neural network generated in advance by machine learning of a plurality of defective images and a plurality of ideal images; a processor that attempts restoration by inputting a read image to the neural network configured using the structure and parameter stored in the storage part and executes decoding processing on the restored read image; and a housing that accommodates a camera, the storage part, and the processor.