Coarse-to-Fine CNN Hierarchy for Multi-Label Classification Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional CNN-based methods for multi-label class classification face challenges in optimizing parameters and accuracy due to the need for prelearning of subnetworks and class grouping, which affects performance.
Innovation Solution
A method and apparatus using a coarse-to-fine convolutional neural network (CF-CNN) with a hierarchical structure, employing a main network and subnetworks, and a disjoint grouping method to generate group labels, allowing for efficient classification without prelearning, and enhancing accuracy through a refine convolutional layer.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the depth of layers and number of filters in CNN are increased to solve nonlinear problems, then classification accuracy is improved, but the number of parameters increases making optimization difficult
Solution Approach 1:
The patent divides the classification task into multiple stages using an ensemble of sub-CNNs, where each sub-CNN handles a specific subset of classes. This segmentation allows each individual network to have fewer parameters while collectively achieving high classification accuracy across all classes.
Solution Approach 2:
The patent employs a hierarchical structure where sub-CNNs are nested within an ensemble framework. Each sub-CNN is a smaller network focused on specific classes, and their outputs are combined to form the final classification, creating a nested architecture that balances complexity and performance.
2Measurement precision
If conventional hierarchical deep CNN algorithm is used to group classes into categories with sub-CNN for each category, then network performance is improved, but the number of subnetworks required increases proportionally with the number of categories requiring prelearning of each subnetwork
Solution Approach 1:
The patent designs sub-CNNs with shared architectural components and weight initialization strategies that allow them to be trained efficiently without requiring extensive prelearning for each individual network. The ensemble framework provides a universal structure that can handle multiple class categories.
Solution Approach 2:
The patent implements a systematic approach where sub-CNNs are trained in a coordinated manner with shared preprocessing and feature extraction pipelines. This preliminary structuring of the training process reduces the need for extensive prelearning of each individual subnetwork.
3Adaptability or versatility
If conventional hierarchical deep CNN algorithm is used with multiple subnetworks, then classification capability is improved, but accuracy is greatly affected by class grouping
Solution Approach 1:
The patent employs dynamic class grouping strategies where the assignment of classes to sub-CNNs can be adjusted based on class characteristics and relationships. This dynamic approach allows the system to adapt the grouping to optimize both classification capability and accuracy, rather than using fixed arbitrary groupings.
Solution Approach 2:
The patent incorporates feedback mechanisms in the training process where the performance of sub-CNNs on different class groupings is evaluated and used to refine the grouping strategy. This feedback loop allows the system to learn optimal class assignments that improve classification accuracy while maintaining versatile classification capability.
Data Source
AI summary
An apparatus for multi-label class classification based on a coarse-to-fine convolutional neural network includes: a processor; and a memory connected to the processor, in which the memory stores program instructions executed by the processor to generate a plurality of hierarchical structure based group labels for a plurality of classes to be classified by using a disjoint grouping method, predict classes which belong to the plurality of group labels, respectively among the plurality of classes by using a coarse-to-fine convolutional neural network including a main network and one or more subnetworks, complete learning of the coarse-to-fine convolutional network through the prediction, and classify one or more classes included in the image by receiving a feature map input from a last convolutional layer of the one or more subnetworks by the main network of the coarse-to-fine convolutional neural network of which learning is completed.