Dimension Fusion for Image Classification Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image classification networks based on deep learning primarily focus on feature extraction and fusion in the image space, ignoring the relationships between features of different dimensions, resulting in poor learning effects and low accuracy.
Innovation Solution
An image processing method that inputs an image into a feature extraction model to generate N-dimensional features, performs dimension fusion on selected M features to obtain M-dimensional fusion features, and processes the image based on both the fusion features and remaining features, considering the relationships between different dimensions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature extraction is performed only in image space using CNN, then the network structure is simple and easy to implement, but the learning effect is poor and accuracy is low due to ignoring relationships between features of different dimensions
Solution Approach 1:
The patent introduces dimension fusion operations that treat feature dimensions as a new spatial dimension. Instead of only processing features in the traditional image space (height, width, channel), the method performs fusion across the dimension axis, creating a multi-dimensional feature processing framework that captures relationships between different feature dimensions while maintaining a relatively simple network structure.
Solution Approach 2:
The patent merges features of different dimensions through dimension fusion operations. Selected features from different dimensions are fused together to create comprehensive feature representations, allowing the network to leverage relationships between dimensions while keeping the overall architecture straightforward and easy to implement.
2Reliability
If all N dimensional features are processed independently, then the processing is simple and fast, but the relationships between different dimensions are ignored resulting in poor learning effects
Solution Approach 1:
The patent applies partial dimension fusion by selecting M features from N dimensional features for fusion, rather than fusing all features. This selective approach captures the essential relationships between dimensions that improve learning effects, while avoiding the excessive complexity that would result from fusing all dimensions. The method strikes a balance between utilizing dimensional relationships and maintaining processing efficiency.
Data Source
AI summary
The present disclosure provides an image processing method. An image to be classified is input into a feature extraction model to generate N dimensional features. Dimension fusion is performed on M features of the N dimensional features to obtain M dimension fusion features. The image to be classified is processed based on M dimension fusion features and remaining features of the N dimensional features other than the M features.