Multi-Int Mediator for Image Quality Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image collection systems rely on manual or rule-based methods, leading to oversubscription and reduced information yield, as they focus on specific products rather than information needs, and lack the ability to suggest alternative collection modalities for image requests, such as converting electro-optical NIIRS Level 5 images to synthetic aperture radar or hyperspectral sensors.
Innovation Solution
Implementing a Multi-Int Mediator (MIM) capable of performing tipping and schedule planning, and using machine learning (ML) to automatically determine image quality based on quantitative metrics like resolution, signal-to-noise ratio, and relative edge response, to filter and prioritize images for analysis tasks, ensuring only sufficient quality images are used for training and inference, thereby enhancing the intelligence yield of the collection system.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual or rule-based methods are used for image collection, then specific product requests can be fulfilled, but resource oversubscription and reduced information yield occur
Solution Approach 1:
The patent introduces a Multi-Int Mediator (MIM) system that acts as an intermediary between image collection requests and the sensor constellation. The MIM performs tipping (suggesting alternative collection modalities) and schedule planning to optimize resource allocation. This mediator transforms manual or rule-based direct requests into intelligent, multi-modal collection strategies, resolving the contradiction by improving information yield while reducing resource congestion through automated mediation.
Solution Approach 2:
The system changes the parameters of image collection by automatically suggesting alternative modalities (e.g., converting electro-optical requests to synthetic aperture radar or hyperspectral sensors). This parameter transformation allows the system to fulfill information needs while distributing load across different sensor types, thereby improving productivity without increasing device complexity.
2Productivity
If images of insufficient quality are used for training and inference, then more images can be processed, but classification errors increase and analysis success rate decreases
Solution Approach 1:
The patent implements preliminary action by automatically determining image quality using machine learning models before images are used for training or inference. The system evaluates quantitative metrics (resolution, signal-to-noise ratio, relative edge response) and NIIRS values in advance, filtering out insufficient quality images before they enter the processing pipeline. This preliminary quality assessment ensures that only sufficient quality images are processed, maintaining high reliability while optimizing productivity.
Solution Approach 2:
The system employs feedback mechanisms where ML models continuously assess image quality and provide feedback on whether images meet the required threshold for specific analysis tasks. This feedback loop enables dynamic filtering and prioritization, ensuring that images used for training and inference meet quality standards, thereby maintaining classification accuracy while managing processing volume effectively.
3Ease of operation
If focus is placed on specific products rather than information needs, then product delivery is straightforward, but alternative collection modalities are not suggested
Solution Approach 1:
The patent introduces dynamics by making the image collection system adaptable and flexible. The Multi-Int Mediator dynamically suggests alternative collection modalities based on information needs rather than rigid product specifications. The system can transform a request for a specific product (e.g., electro-optical image) into alternative modalities (synthetic aperture radar, hyperspectral) when appropriate, enabling the system to adapt to varying conditions and optimize information yield while maintaining ease of operation through automated decision-making.
Data Source
AI summary
Discussed herein are architectures and techniques for improving execution or training of machine learning techniques. A method can include receiving a request for image data, the request indicating an analysis task to be performed using the requested image data, determining a minimum image quality score for performing the analysis task, issuing a request for image data associated with an image quality at last equal to, or greater than, the determined minimum image quality score, receiving, in response to the request, image data with an image quality score greater than, or equal to, the determined minimum image quality score, and providing the received image data to (a) a machine learning (ML) model executor to perform the image analysis task or (b) an ML model trainer that trains the ML model to perform the image analysis task.