Sheet Identification Using Machine Learning Partial Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sheet type identification methods using optical sensors are prone to inaccuracies due to individual sensor differences and degradation over time, leading to insufficient identification accuracy, especially when sheet types have similar characteristics.
Innovation Solution
A machine learning-based identification system that employs an estimation model with multiple partial models to process parameters from sensors, distinguishing between values influenced by sensor degradation and those unaffected, allowing for improved sheet type classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If sheet type identification is performed based on collation between a value specified in advance and a value of reflected light/transmitted light obtained using an optical sensor, then the identification process is simple, but identification accuracy deteriorates due to individual sensor differences and degradation over time
Solution Approach 1:
The system performs preliminary learning to create a determination model that accounts for individual sensor characteristics and degradation patterns. By pre-processing sensor data through machine learning before actual identification, the system adapts to each sensor's unique characteristics in advance, thereby maintaining high accuracy without complicating the operational identification process.
Solution Approach 2:
The invention transforms the identification approach by changing from direct value collation to a machine learning-based determination model. The model learns optimal parameter relationships and degradation patterns from training data, dynamically adjusting identification criteria to compensate for sensor variations and aging, thus improving accuracy while keeping the user interface simple.
2Device complexity
If a single determination model is used for all sensors, then the system structure is simple, but identification accuracy deteriorates due to individual sensor differences
Solution Approach 1:
The determination model performs self-adjustment by learning individual sensor characteristics during a preliminary learning phase. Each sensor's unique response patterns and degradation behaviors are automatically captured and compensated for in the model, enabling the system to adapt to individual sensor differences without requiring manual calibration or complex hardware modifications.
Solution Approach 2:
A preliminary learning process is executed to train the determination model with data from specific sensors before actual identification operations. This pre-training phase allows the model to internalize each sensor's characteristics, ensuring high accuracy from the start of normal operation without adding structural complexity to the identification system.
3Productivity
If conventional identification methods are used, then processing speed is fast, but identification accuracy deteriorates when sheet types have similar characteristics
Solution Approach 1:
The system performs preliminary learning to build a sophisticated determination model that captures subtle differences between similar sheet types. By pre-processing and learning from comprehensive training data, the model develops nuanced understanding of sheet characteristics, enabling accurate differentiation of similar types during fast operational identification without compromising speed.
Solution Approach 2:
The invention transitions from simple threshold-based collation to a machine learning determination model that analyzes multiple parameters and their relationships. This parameter transformation approach enables the system to detect subtle differences between similar sheet types by leveraging learned patterns, maintaining high processing speed through efficient model inference while dramatically improving accuracy for difficult distinctions.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enhances sheet type identification accuracy by accounting for sensor degradation and individual differences, enabling precise classification of sheet types with similar characteristics.
Implementation Method 1
the sheet is illuminated with light, and sheet type identification is performed based on output values obtained when a sensor receives reflected light and transmitted light
Data Source
AI summary
An identification apparatus for identifying a type of a sheet obtains, by a sensor, a plurality of parameters including a parameter concerning a characteristic of the sheet, and identifies the type of the sheet based on a result of inputting the plurality of parameters obtained by the sensor to an estimation model obtained by machine learning using, as input data, parameters corresponding to the parameters which are obtained by the sensor and include a first parameter and a second parameter classified in accordance with whether a value tends to change due to a predetermined element, and also using, as supervised data, a type of a sheet if the parameters are obtained. The estimation model includes a first partial model to which the first parameter is input and the second parameter is not input, a second partial model to which the second parameter is input and the first parameter is not input, and a third partial model that outputs a result concerning the type of the sheet based on an output from the first partial model and an output from the second partial model.