Image–Scalar Data Fusion for Lower-Complexity Sensor Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for analyzing sensor data, particularly those involving neural networks, are complex, time-consuming, and require large amounts of training data, which are often not available in sufficient quantities.
Innovation Solution
A method that combines image information and scalar sensor values into consistent data structures for comparison, allowing for more complex data analysis without relying solely on neural networks, optionally using a simpler neural network for verification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Difficulty of detecting and measuring
If neural networks are used to analyze sensor data, then analysis capability is improved, but system complexity and training requirements increase
Solution Approach 1:
The patent segments the analysis system into two parts: a neural network component for learning temporal patterns from time-series sensor data, and a traditional image processing component for analyzing spatial information. This segmentation allows each component to specialize in its strength while avoiding the complexity of using a single complex neural network for all analysis tasks.
Solution Approach 2:
The patent merges neural network output (temporal pattern recognition) with traditional image processing results (spatial analysis) to create a comprehensive analysis system. By combining these two approaches, the system achieves high analysis capability without requiring a single monolithic complex neural network.
2Measurement precision
If neural networks are trained to improve analysis accuracy, then measurement precision is improved, but training time and data requirements increase
Solution Approach 1:
The patent extracts only the temporal pattern recognition function into the neural network, while leaving spatial analysis to traditional image processing methods. This extraction allows the neural network to be trained on a smaller, more focused dataset for temporal patterns, reducing training time and data requirements while maintaining high analysis accuracy through specialized processing.
Solution Approach 2:
Instead of using a neural network for complete analysis, the patent applies partial neural network processing only where it adds value (temporal pattern recognition). This partial application reduces the overall training burden while maintaining sufficient accuracy for the specific analysis tasks.
3Reliability
If more training data is collected to improve neural network reliability, then system reliability is improved, but data collection time and resources increase
Solution Approach 1:
The patent extracts the temporal analysis function from the overall system and handles it with a neural network that requires less training data, while spatial analysis uses traditional methods that don't require extensive training. This extraction reduces the total amount of training data needed while maintaining system reliability through complementary analysis approaches.
Solution Approach 2:
The patent changes the approach parameters by using different analysis methods for different data dimensions (temporal vs. spatial). This parameter change allows the system to achieve reliability through methodological diversity rather than relying solely on large volumes of training data for a single approach.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention relates to a method for analysing image information (134, 234) using assigned scalar values (119, 129, 219, 229), comprising the method steps of: a.) capturing a first image (134) of an object (132, 232) or a situation (130) and capturing at least one first scalar sensor value (129) relating to the object (132) or the situation (130); b.) capturing a second image (234) of the object (132, 232) or the situation (130) and capturing at least one second scalar sensor value (229) relating to the object (132, 232) or the situation (130); c.) inserting the first image (134) and the at least one first scalar sensor value (129) into a first data structure (159) as a consistent representation form, and inserting the second image (234) and the at least one second scalar sensor value (229) into a second data structure (259) as a consistent representation form; d.) comparing the first (159) and the second data structure (259); and e.) outputting information if the comparison results in a difference that corresponds to a defined or definable criterion.