Automated Display Verification Using Siamese CNN Feature Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual verification of display software for avionics systems is time-consuming and prone to human error, particularly when dealing with complex images containing multiple layers and varying backgrounds, which delays deployment and compliance with regulatory requirements.
Innovation Solution
An automated validation system using a Siamese configuration of a convolutional neural network (CNN) to analyze unverified images and reference images, employing kernel shortlisting criteria to extract feature maps and calculate a similarity score, thereby validating images based on the similarity of graphical components and layout without human intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual verification is used to ensure compliance with regulatory requirements, then measurement precision and reliability are improved, but time consumption increases and productivity decreases
Solution Approach 1:
The patent replaces manual visual verification with an automated computer vision system using convolutional neural networks. The system captures display images, extracts features using CNN layers, and automatically compares them against reference images to determine compliance, eliminating the need for manual inspection while maintaining verification reliability.
Solution Approach 2:
The verification system performs self-validation by automatically comparing generated display images against stored reference images that represent compliant displays. The system independently identifies features, extracts characteristics, and determines compliance without requiring human intervention in the verification process itself.
2Measurement precision
If manual testing is performed to verify display compliance, then measurement precision is improved, but loss of time increases
Solution Approach 1:
The patent substitutes manual visual inspection with an automated image processing system that uses convolutional neural networks to extract features and compare display images against reference images, achieving both high measurement precision and reduced verification time through computational analysis.
Solution Approach 2:
The system performs preliminary feature extraction and reference image storage before actual verification occurs. Reference images representing compliant displays are pre-stored in a database, and the verification process simply involves comparing new images against these pre-prepared references, significantly reducing verification time while maintaining precision.
3Productivity
If automated verification is implemented to reduce time to market, then productivity is improved, but device complexity increases
Solution Approach 1:
The verification system is segmented into distinct functional modules: image capture module, feature extraction module using CNN layers, comparison module that contrasts extracted features against reference images, and compliance determination module. This segmentation allows each component to perform its specific function independently, managing overall system complexity while achieving high productivity.
Data Source
AI summary
Methods and systems are provided for automated validation of images generated by uncertified software. One method involves analyzing an unverified image and a reference image in parallel to identify sets of kernels associated with a layer of a convolutional neural network, creating a shared set of kernels associated with that layer in accordance with one or more kernel shortlisting criteria, obtaining a first feature map for the unverified image using that layer of the convolutional neural network and the shared set of kernels, obtaining a second feature map for the reference image using that layer of the convolutional neural network and the shared set of kernels, calculating a similarity score for the unverified image based at least in part on differences between the feature maps and a weighting factor associated with the first layer, and automatically validating the unverified image based at least in part on the similarity score.