ANN Image Classifier Using Class-Specific Feature Segregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional feature extraction techniques in machine learning and deep learning applications, such as image classification, lack class-specific information, leading to sub-par performance and increased computations due to the generic and reduced nature of the features extracted.
Innovation Solution
A method and system for updating an artificial neural network (ANN) based image classifier by determining relative relevance values of features for each class, segregating co-occurring features, identifying imbalances, and adjusting weights to balance class-specific relevant features, thereby enhancing the classifier's performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional feature extraction techniques are used to reduce dimensionality of input data, then processing efficiency is improved, but class-specific information is lost leading to reduced classification accuracy
Solution Approach 1:
The patent segments features into class-specific relevant features and generic features. By separating features that are specific to each class from generic features, the system maintains class-specific information while still achieving dimensionality reduction through selective use of only the necessary class-specific features for each classification task.
Solution Approach 2:
The patent applies local quality by providing different feature sets tailored to each specific class rather than using a single generic feature set for all classes. Each class receives features that are locally optimized for its specific characteristics, improving classification accuracy while maintaining efficiency by not processing unnecessary features.
2Loss of time
If conventional feature extraction techniques provide reduced and generic features, then computation time is reduced, but performance of classification models deteriorates
Solution Approach 1:
The patent extracts and isolates only the relevant class-specific features needed for each classification task, removing unnecessary generic features. This extraction process maintains computation efficiency by reducing feature dimensionality while preserving the essential class-specific information required for accurate classification.
Solution Approach 2:
The patent changes the parameters of feature representation by identifying and using only the most relevant features for each class. By dynamically selecting and weighting features based on their class-specific relevance, the system achieves both computational efficiency and high classification performance.
3Device complexity
If generic reduced features are used across all classes, then system complexity is reduced, but accuracy for specific class classification deteriorates
Solution Approach 1:
The patent segments the feature space into class-specific components, creating dedicated feature representations for each class. This segmentation allows the system to maintain relatively simple processing for each individual class while achieving high accuracy through class-specific feature optimization.
Solution Approach 2:
The patent introduces dynamics by adapting the feature set to each specific classification task. Rather than using a static generic feature set, the system dynamically selects and weights features based on their relevance to each particular class, achieving high accuracy without requiring a permanently complex system structure.
Data Source
AI summary
The disclosure relates to method and system for training an artificial neural network (ANN) based image classifier using class-specific relevant features. The method includes receiving the ANN based image classifier, training image dataset, and various features of the training image dataset. The method further includes determining a relative relevance value of each of the features corresponding to each of the classes based on the ANN based image classifier, segregating co-occurring features from the features for each of the classes based on the training image dataset and the ANN based image classifier, identifying an imbalance in the class-specific relevant features for each of the classes based on the relative relevance value of each of the features corresponding to each of the classes, and updating the ANN based image classifier based on the imbalance in the class-specific relevant features and the co-occurring features for each of the classes.