Pattern Recognition Reference Data Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional pattern recognition systems face challenges in efficiently learning and updating reference data, leading to increased processing time and difficulties in integrating power consumption and circuit size, especially when recognizing new patterns or adding new reference data.
Innovation Solution
A reference data optimization learning method that calculates distances between input data and reference data, optimizes the position of reference data to a gravity center, and adjusts threshold values to prevent overlap, allowing for efficient learning and recognition of new reference data within a short time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If sequential comparison method is used for pattern recognition, then processing is simple and hardware requirements are low, but processing time increases in proportion to the number of comparison data
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing distance values between reference data and input data before actual pattern recognition. The distance calculation unit computes distances in advance, and these pre-computed distances are stored in memory for quick retrieval during recognition, eliminating the need for real-time sequential comparison and significantly reducing processing time.
Solution Approach 2:
The patent introduces an intermediary mechanism by using a distance calculation unit and memory storage as intermediaries between the input data and the pattern recognition process. Instead of directly comparing input patterns with reference data, the system uses pre-calculated distance values as intermediaries to determine pattern matches, thereby reducing the computational burden during actual recognition.
2Adaptability or versatility
If neural network method is used for pattern recognition, then learning capability is improved, but threshold value updates and network load updates require great deal of processing time
Solution Approach 1:
The patent segments the neural network update process into independent distance calculation operations. Instead of updating all threshold values and network loads simultaneously as a monolithic process, the system calculates distances between individual reference data points and input data separately, allowing for more efficient and parallelizable updates that reduce overall processing time.
Solution Approach 2:
The patent replaces the traditional mechanical neural network update mechanism with a distance-based calculation system. Instead of iteratively adjusting threshold values and network weights through complex feedback loops, the system directly computes distances between data points and uses these distances to determine pattern recognition outcomes, significantly simplifying the update process.
3Reliability
If entire network relearning is performed to recognize new pattern, then recognition accuracy is maintained, but processing time and power consumption increase
Solution Approach 1:
The patent extracts only the necessary components for recognizing new patterns by calculating distances specifically between the new input data and existing reference data. Instead of relearning the entire network, the system extracts and processes only the relevant distance calculations needed for the new pattern, maintaining recognition accuracy while minimizing processing time and power consumption.
4Speed
If associative memory with fully parallel architecture is used, then pattern matching speed is improved, but circuit size and power consumption integration becomes difficult
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing distance values in memory before actual pattern matching occurs. This allows the system to use simpler, more integrable circuitry during the matching phase, as the complex distance calculations have already been performed and stored, reducing the need for complex parallel processing circuits while maintaining high-speed matching capability.
Data Source
AI summary
The present invention is directed to a pattern recognition system in which new reference data to be added is efficiently learned. In the pattern recognition system, there is performed the calculation of distances equivalent to similarities between input data of a pattern search target and a plurality of reference data, and based on input data of a fixed number of times corresponding to the reference data set as a recognized winner, a gravity center thereof is calculated to optimize the reference data. Furthermore, a threshold value is changed to enlarge/reduce recognition areas, whereby erroneous recognition is prevented and a recognition rate is improved.