Thermal Image Classification for System Performance Optimization

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing systems face challenges in categorizing amorphous or featureless images, such as computer-generated heatmaps, which are difficult for machines to understand without prior tagging, and require improved classification processes for system performance optimization.

Innovation Solution

A computing platform trains a thermal image classification model using historical system performance data to generate and classify thermal images, identifying correlations between image features and system performance, and dynamically adjusts weighting values to optimize classification accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional machine learning methods are used to categorize images, then images with distinct features and boundaries can be easily classified, but featureless or amorphous images cannot be effectively categorized

Engineering Contradiction:

Improveimage classification accuracyVSAvoidapplicability to featureless images

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent transforms the classification approach by changing parameters from feature-based analysis to structural property-based analysis. It extracts structural properties (edges, contours, spatial frequency, entropy) from thermal images and uses these as classification parameters, enabling effective categorization of featureless images that traditional methods cannot handle.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent introduces structural property analysis as an intermediary between the thermal image and the classification model. By extracting intermediate structural features (edges, contours, spatial frequency patterns) and using them to train the classification model, the system bridges the gap between featureless images and meaningful categorization.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Measurement precision

If a classification model is trained on historical thermal images to improve classification accuracy, then system performance monitoring improves, but computational resources and training time increase

Engineering Contradiction:

Improveclassification accuracyVSAvoidmodel training time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary action by pre-training the classification model using historical thermal images and their corresponding system performance data. This offline training phase prepares the model in advance, so that during runtime, classification can be performed quickly without requiring real-time training, thus reducing operational time loss.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent implements a dynamic feedback loop where the classification model is continuously retrained and updated based on new thermal images and system performance data. This dynamic updating allows the model to adapt to changing system conditions and improve accuracy over time without requiring complete retraining, balancing accuracy improvement with time efficiency.

Inventive Principle:

Principle #15Dynamics

3Measurement precision

If dynamic feedback loop is used to update weighting values in the classification model, then classification accuracy improves over time, but system complexity increases

Engineering Contradiction:

Improveclassification accuracyVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent implements a dynamic feedback loop where classification results and system performance data are fed back to update the weighting values of structural properties in the classification model. This feedback mechanism automatically adjusts the model parameters based on real-world performance, improving accuracy without requiring manual intervention or complex reconfiguration.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The classification model performs self-service by automatically updating its own weighting values through the feedback loop. The system uses its own classification outputs and corresponding system performance data to self-adjust and optimize its parameters, reducing the need for external intervention and simplifying the overall system architecture.

Inventive Principle:

Principle #25Self-service

Data Source

PatentUS20240354596A1Featureless Image Categorization and Recognition using Hybrid Artificial Intelligence (AI) and Image Structural Properties for System Performance Optimization

Publication Date: 2024.10.24 BANK OF AMERICA CORP

AI summary

A computing platform may train, using a plurality of historical thermal images, a thermal image classification model, which may configure the thermal image classification model to classify thermal images based on performance of systems represented by the thermal images. The computing platform may collect current system performance information for a first computing system. The computing platform may generate, using the current system performance information, a new thermal image, representative of the current system performance information. The computing platform may classify, using the thermal image classification model, the new thermal image. Based on the classification of the new thermal image, the computing platform may send one or more network action commands, which may cause a network traffic manager to redirect traffic from the first computing system to a second computing system.