RGB–Hyperspectral Classification for Camouflaged Object Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing object recognition techniques using RGB images fail in environments with insufficient contrast, particularly in military settings where objects are camouflaged, leading to inaccurate detection.

Innovation Solution

A computer-implemented method that combines RGB and hyperspectral image classification using multiple classifier models to generate object classifications, incorporating both color and material information, and amalgamates these classifications to improve detection accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If RGB image classification is used for object detection, then the detection process is simple and fast, but detection accuracy deteriorates in camouflaged environments with insufficient contrast

Engineering Contradiction:
Improvedetection speedVSAvoiddetection accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent combines RGB image data with hyperspectral image data to create a multi-modal classification system. The RGB classifier and hyperspectral classifier work together, merging their respective strengths to achieve both speed and accuracy in object detection, particularly in camouflaged environments where single-modality approaches fail.

Inventive Principle:
Principle #5Merging (Combining)

2Measurement precision

If hyperspectral image classification is used for object detection, then detection accuracy improves in camouflaged environments, but device complexity and processing requirements increase

Engineering Contradiction:
Improvedetection accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the classification task into two independent parts: an RGB-based classifier for rapid initial assessment and a hyperspectral-based classifier for detailed material analysis. This segmentation allows the system to leverage hyperspectral data for improved accuracy without requiring the entire system to be complex, as each classifier can be optimized independently for its specific data type.

Inventive Principle:
Principle #1Segmentation

3Measurement precision

If multiple classifier models are used to amalgamate classifications, then object classification accuracy improves, but computational complexity increases

Engineering Contradiction:
Improveclassification accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces an intermediary amalgamation module that combines the outputs of the RGB classifier and hyperspectral classifier. This intermediary component integrates the classifications from both models, allowing the system to benefit from multiple classification perspectives while managing computational complexity through a structured combination approach rather than requiring all models to operate simultaneously at full complexity.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4645248A1Further improvements in classifying an object
Publication Date: 2025.11.05 BAE SYSTEMS PLC
  • EP4645248A1 patent drawingFigure 1
  • EP4645248A1 patent drawingFigure 2~4
  • EP4645248A1 patent drawingFigure 3

AI summary

The present disclosure relates to a computer-implemented method of classifying an object from a real-world environment in which an entity operates, the computer-implemented method comprises: receiving an image from a camera mounted to the entity and a hyperspectral image from a hyperspectral camera mounted to the entity; obtaining a plurality of object classifications from a plurality of different classifier models, respectively; and generating an object classification by amalgamating the plurality of object classifications, wherein the plurality of different classifier models includes at least one classifier model configured to generate an object classification using the image and at least one classifier model configured to generate an object classification using the hyperspectral image