Cross-Domain Image Learning with Reference-Angle Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In image classification, particularly in detailed classification where variance within categories is large and variance between categories is small, increasing training data through data augmentation is challenging due to the reliance on local features, making accurate classification difficult.
Innovation Solution
A learning device and method that includes data acquisition, alignment, feature extraction, and classification of local features from both source and target domains, followed by training to align and normalize these features using a reference angle, enabling domain adaptation to improve classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If data augmentation is used to increase training data amount, then the amount of training data is increased, but classification accuracy cannot be improved in detailed classification problems because local features are used
Solution Approach 1:
The patent introduces alignment processing as an intermediary step between data acquisition and feature extraction. The alignment processing converts images from different domains (source domain and target domain) to a common reference domain, enabling effective feature extraction and classification. This intermediary alignment step resolves the contradiction by making the data from different domains comparable and usable for accurate classification.
Solution Approach 2:
The patent changes the parameter domain by performing domain adaptation through alignment processing. Images are transformed from their original domain parameters to a reference domain, effectively changing the parameter space. This allows the model to learn from source domain data and apply it to target domain data, improving classification accuracy without requiring abundant target domain training data.
2Measurement precision
If local features are extracted for detailed classification, then classification of similar categories becomes possible, but data augmentation cannot be simply applied to increase learning data
Solution Approach 1:
The alignment processing acts as an intermediary that enables data augmentation to work with local features. By first aligning images to a reference domain and then extracting local features, the system can effectively augment data while maintaining the precision needed for detailed classification of similar categories.
Solution Approach 2:
The patent performs alignment processing as a preliminary action before feature extraction and classification. This preliminary alignment ensures that subsequent local feature extraction and data augmentation operations are performed on properly normalized data, enabling the system to achieve both high precision for similar categories and effective data augmentation.
3Reliability
If source domain data and target domain data are used for training, then model robustness is improved, but domain difference causes classification accuracy to deteriorate
Solution Approach 1:
The alignment processing serves as an intermediary that bridges source domain and target domain data. By converting both domains to a reference domain, the system can train on diverse data (improving robustness) while eliminating domain-specific variations that would otherwise harm classification accuracy.
Solution Approach 2:
The patent creates equipotentiality between different domains by aligning them to a common reference domain. This normalization ensures that data from source and target domains are on the same level, allowing the model to learn from both without the domain differences causing accuracy deterioration.
Data Source
AI summary
A data acquisition means acquires source domain data and target domain data. An alignment means performs an alignment which converts the source domain data and the target domain data into images of a predetermined reference angle. A feature extraction means extracts local features of the source domain data and the target domain data. A classification means classifies a class based on the local features of the source domain data and the target domain data after the alignment. A learning means trains the feature extraction means based on the local features of the source domain data and the target domain data after the alignment and a classification result of the class.