BCI Image Classification via Segmented Linear-Nonlinear Pipeline

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing Brain Computer Interface (BCI) systems face challenges in efficiently classifying images in real-time due to the complexity of non-linear classifiers and their susceptibility to overfitting, which hinders their performance in practical applications.

Innovation Solution

The proposed method and system utilize a combination of computer vision techniques and neurophysiological signal processing to classify images. This involves applying computer vision procedures to detect potential image regions, presenting these regions to an observer, recording their brain responses using EEG, and processing these signals to identify neurophysiological events indicative of target detection. The system employs classifiers such as Spatially Weighted Fisher Linear Discriminant (SWFLD) and convolutional neural networks (CNN) to integrate both computer vision and neurophysiological data for accurate image classification.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If non-linear classifiers such as Neural Networks are used to capture complex discriminative functions, then classification accuracy is improved, but device complexity and computational demands increase significantly

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

Solution Approach 1:

The patent segments the classification process into two distinct stages: (1) a linear classification stage using methods like LDA or SVM to obtain initial class labels and discriminative features, and (2) a non-linear refinement stage using neural networks only on the reduced feature set from stage 1. This segmentation allows the system to benefit from both linear and non-linear approaches while avoiding the full computational burden of applying non-linear classifiers to the complete original feature space.

Inventive Principle:
Principle #1Segmentation

2Measurement precision

If non-linear classifiers are used to capture complex discriminative functions, then classification accuracy is improved, but susceptibility to overfitting increases

Engineering Contradiction:
Improveclassification accuracyVSAvoidrobustness to overfitting
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent segments the classification process into two distinct stages: (1) a linear classification stage using methods like LDA or SVM to obtain initial class labels and discriminative features, and (2) a non-linear refinement stage using neural networks only on the reduced feature set from stage 1. This segmentation allows the system to benefit from both linear and non-linear approaches while avoiding the full computational burden of applying non-linear classifiers to the complete original feature space.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts and removes redundant features from the original high-dimensional feature space through linear classification in the first stage. By taking out only the most discriminative features and using them as input for the non-linear neural network classifier in the second stage, the system reduces the risk of overfitting while maintaining the ability to capture complex patterns.

Inventive Principle:
Principle #2Taking out (Extraction)

3Device complexity

If linear classifiers are used to reduce complexity and avoid overfitting, then device complexity is reduced, but ability to capture complex discriminative functions is limited

Engineering Contradiction:
Improveclassifier complexityVSAvoidclassification accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent segments the classification process into two distinct stages: (1) a linear classification stage using methods like LDA or SVM to obtain initial class labels and discriminative features, and (2) a non-linear refinement stage using neural networks only on the reduced feature set from stage 1. This segmentation allows the system to benefit from both linear and non-linear approaches while avoiding the full computational burden of applying non-linear classifiers to the complete original feature space.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent merges the strengths of linear and non-linear classifiers by combining them in a two-stage pipeline. The first stage uses linear classifiers to reduce dimensionality and provide initial classification, while the second stage incorporates non-linear neural networks to refine the classification and capture complex patterns that linear methods miss.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP3304255B1Image classification by brain computer interface
Publication Date: 2025.05.07 INNEREYE
  • EP3304255B1 patent drawingFigure 1A
  • EP3304255B1 patent drawingFigure 1B
  • EP3304255B1 patent drawingFigure 1C

AI summary

A method of classifying an image is disclosed. The method comprises: applying a computer vision procedure to the image to detect therein candidate image regions suspected as being occupied by a target; presenting to an observer each candidate image region as a visual stimulus, while collecting neurophysiological signals from a brain of the observer; processing the neurophysiological signals to identify a neurophysiological event indicative of a detection of the target by the observer; and determining an existence of the target in the image is based, at least in part, on the identification of the neurophysiological event.