Neural Network Recognition via Interconnected Feature Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neural network recognition methods face challenges in efficiently processing complex patterns and improving recognition rates, particularly in tasks that require multiple classifications, due to limitations in feature map computation and training processes.
Innovation Solution
The proposed method involves a neural network structure with multiple interconnected layers, where element-wise operations are applied to feature maps from different neural networks to generate intermediate feature maps, enabling improved recognition results by preprocessing and transmitting feature maps between networks, and training networks based on recognition results and outputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single neural network is used for recognition, then the device complexity is low, but the recognition accuracy for complex patterns is insufficient
Solution Approach 1:
The patent combines multiple neural networks into a unified architecture where feature maps from different networks are integrated through element-wise operations. This merging approach maintains the computational benefits of multiple specialized networks while reducing overall system complexity through shared structures and operations.
Solution Approach 2:
The neural network architecture is designed to perform multiple recognition tasks simultaneously by processing different input types (e.g., color, texture, shape) through shared layers. This multi-functionality allows a single device to handle complex pattern recognition across various modalities without requiring separate dedicated networks for each task.
2Measurement precision
If multiple neural networks are used to process complex patterns, then the recognition accuracy improves, but the memory size increases
Solution Approach 1:
Multiple neural networks are merged into a shared architecture where common layers and operations are consolidated. This reduces redundant memory storage for weights and biases while maintaining the computational capabilities of processing multiple feature types simultaneously through element-wise operations on shared feature maps.
3Measurement precision
If traditional feature map computation is used, then the processing speed is fast, but the recognition rate for complex patterns is limited
Solution Approach 1:
The system performs preliminary feature extraction through element-wise operations on feature maps from multiple neural networks before final recognition decisions are made. This preliminary processing enriches the feature representation with multiple modalities (color, texture, shape) in advance, enabling faster and more accurate recognition without requiring complex post-processing computations.
Data Source
AI summary
A neural network recognition method includes obtaining a first neural network that includes layers and a second neural network that includes a layer connected to the first neural network, actuating a processor to compute a first feature map from input data based on a layer of the first neural network, compute a second feature map from the input data based on the layer connected to the first neural network in the second neural network, and generate a recognition result based on the first neural network from an intermediate feature map computed by applying an element-wise operation to the first feature map and the second feature map.