Convolutional Neural Network Feature Map Decomposition for Accuracy and Overhead
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Convolutional neural networks require excessive processing overhead and may produce inaccurate classifications, which can lead to inefficiencies and safety issues in applications like predictive maintenance, cybersecurity, and autonomous vehicles.
Innovation Solution
The network decomposes feature maps into high and low frequency components, using learned upsampling techniques to combine them, reducing the number of convolutions needed for feature extraction and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional convolutional neural networks are used for feature extraction, then classification can be performed, but processing overhead is excessive and accuracy is insufficient
Solution Approach 1:
The patent segments the feature extraction process into two distinct pathways: a first convolutional layer that extracts features from original feature maps, and a second convolutional layer that extracts features from upsampled feature maps. This segmentation allows the network to process different frequency components separately and combine them, improving classification accuracy while managing processing overhead through structured decomposition
Solution Approach 2:
The patent introduces upsampling operations that transform feature maps to different dimensional representations. By creating upsampled feature maps with increased resolution and combining them with original feature maps in a multi-dimensional feature extraction space, the network captures both fine-grained and coarse-grained features, thereby improving classification accuracy
2Measurement precision
If more convolutions are performed to improve accuracy, then classification precision increases, but processing time increases
Solution Approach 1:
The patent performs upsampling of feature maps before the second convolutional layer processing. This preliminary action prepares the feature maps in advance, allowing the second convolutional layer to work with preprocessed data that contains both high-frequency and low-frequency information, thereby improving accuracy without requiring additional extensive processing steps later
Solution Approach 2:
The patent merges features from two different processing pathways: features extracted from original feature maps by the first convolutional layer and features extracted from upsampled feature maps by the second convolutional layer. This merging of feature representations in a combined feature space achieves superior classification accuracy while avoiding the need for excessively deep single-pathway processing that would increase processing time
Data Source
AI summary
Improved convolutional neural network-based machine learning models are disclosed herein. A convolutional neural network is configured to decompose feature maps generated based on a data item to be classified. The feature maps are decomposed into a first and second subsets. The first subset is representative of high frequency components of the data item, and the second subset is representative of low frequency components of the data item. The second subset is upsampled and is combined with the first subset. The combined feature maps are convolved with a filter to extract a set of features associated with the data item. The first subset is also downsampled and combined with the second subset. The combined feature maps are convolved with a filter to extract another set of features. The data item is classified based on the sets of features extracted based on the convolution operations.