Hierarchical Address Recognition Beyond OCR for Faster Mail Sorting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems for items, such as articles of mail, are limited by the time and resource intensity of optical character recognition (OCR) processes when handling large volumes of items, particularly in sorting and delivery operations.
Innovation Solution
A system and method utilizing machine learning or deep learning models to recognize geographical area information, such as addresses, by building hierarchical data sets and sequentially processing geographical area components without relying on OCR, significantly reducing processing time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional OCR processes are used to recognize geographical area information on items, then recognition accuracy can be maintained, but processing time increases significantly (100 milliseconds per item)
Solution Approach 1:
The patent segments the geographical area information recognition task into hierarchical components (state, city, county, township, street, number) and processes each component separately using dedicated machine learning models. This segmentation allows parallel processing of different address components, reducing overall processing time while maintaining recognition accuracy through specialized models for each component type.
Solution Approach 2:
The patent replaces the traditional OCR mechanical character recognition system with machine learning-based image recognition models. Instead of using OCR to convert images of text into machine-encoded text, the system uses trained neural networks to directly recognize and classify geographical area components from images, achieving faster processing (2 milliseconds per item) while maintaining or improving recognition accuracy.
2Productivity
If machine learning models are used to recognize geographical area information, then processing speed increases dramatically (2 milliseconds per item), but system complexity increases
Solution Approach 1:
The system divides the complex address recognition task into manageable hierarchical segments (state, city, county, township, street, number), with each segment processed by a specialized machine learning model. This segmentation reduces the complexity of individual models while maintaining overall system accuracy, and enables parallel processing to achieve high productivity.
Solution Approach 2:
The patent creates a universal hierarchical framework that can process various types of geographical area information across different formats and styles. The multi-level hierarchical models serve multiple functions: classification, recognition, and validation of address components, reducing system complexity by consolidating these functions into a unified architecture.
3Productivity
If hierarchical multi-level models are used to process geographical area components, then sorting efficiency improves, but computational resources required increase
Solution Approach 1:
The hierarchical model processes address components in a structured sequence from larger geographical areas (state) to smaller ones (number), allowing early filtering and validation at each level. This segmentation enables the system to terminate processing early for clearly identifiable addresses, reducing computational resource consumption while maintaining high sorting efficiency.
Solution Approach 2:
The system performs preliminary classification and validation at each hierarchical level before proceeding to more detailed recognition. This preliminary action at state, city, and county levels prepares the data for subsequent processing, reducing the computational burden on later stages and improving overall sorting efficiency with optimized resource usage.
Data Source
AI summary
This application relates to a system for automatically recognizing geographical area information provided on an item. The system may include an optical scanner configured to capture geographical area information provided on an item, the geographical area information comprising a plurality of geographical area components. The system may also include a controller in data communication with the optical scanner and configured to recognize the captured geographical area information by running a plurality of machine learning or deep learning models separately and sequentially on the plurality of geographical area components of the captured geographical area information.