Learning Apparatus Using Subset Classifiers for Image Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image identification technologies face challenges in accurately identifying objects using sensor information, as sensor data from different sources often lack correlation, making it difficult to estimate sensor information effectively.
Innovation Solution
A learning apparatus that acquires and processes multiple image datasets along with sensor information, creating subsets with varying combinations of images and sensor data, and uses a k-division cross verification method to learn and integrate classifiers for accurate image identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sensor information from multiple sources is integrated for image identification, then identification accuracy is improved, but the complexity of data processing increases
Solution Approach 1:
The patent segments the learning process by creating multiple subsets from the learning data set, where each subset contains different combinations of images and sensor information. Multiple subset classifiers are learned independently on these subsets, and their results are integrated by a selection unit. This segmentation approach enables the system to handle multiple sensor sources while maintaining manageable processing complexity for each individual classifier.
2Reliability
If multiple subset classifiers are learned and integrated, then classification reliability is improved, but the learning time and computational resources increase
Solution Approach 1:
The patent employs k-fold cross-validation to create subsets, where each subset uses a portion of the available data. The selection unit integrates results from multiple subset classifiers, selecting the most reliable classification outcome. This approach achieves improved reliability through multiple classifiers while controlling learning time by using cross-validation to efficiently distribute the learning workload across different data partitions.
3Quantity of substance
If sensor information is estimated from other sensor information, then data completeness is improved, but estimation accuracy deteriorates when correlation is weak
Solution Approach 1:
The patent uses image information as an intermediary to bridge sensor information that may have weak direct correlation. The learning apparatus learns classifiers that integrate both image data and sensor data, allowing the system to estimate or infer sensor information characteristics through the mediating role of image data. This intermediary approach enables data completeness improvement while maintaining estimation accuracy by using the image as a common reference frame.
Data Source
AI summary
A learning apparatus includes an acquisition unit, a creation unit, and a first learning unit. The acquisition unit acquires a plurality of leaning data sets including a plurality of images imaged by a plurality of imaging devices, and sensor information of the imaging devices when the plurality of respective images is imaged. The creation unit creates, from the plurality of the plurality of learning data sets, a plurality of subsets, wherein each of the plurality of subsets has a different combination of the plurality of images and the sensor information. The first learning unit learns a plurality of first classifiers respectively corresponding to the plurality of subsets based on the plurality of respective subsets.