TensorCSBP: A Tensor Center-Symmetric Feature Extractor for EEG Odor Detection

TL;DR

This paper introduces TensorCSBP, a new method for extracting features from EEG signals to detect odors with high accuracy and interpretability.

Contribution

TensorCSBP is a novel tensor-based feature extractor that improves explainability and performance in EEG odor detection.

Findings

TensorCSBP achieved 96.68% accuracy in odor detection using a 32-channel EEG dataset.

The DLob method produced a symbol sequence with information entropy of 3.5675, showing rich interpretability.

Abstract

Objective: Accurate odor classification from EEG signals requires informative and interpretable features. Although Local Binary Pattern (LBP) and variants such as the center-symmetric binary pattern are widely used, they lack sufficient explainability and tensor-level implementations. Additionally, neuroscientific understanding of odor processing remains limited. Methods: We propose Tensor Center-Symmetric Binary Pattern (TensorCSBP), a novel tensor-based feature extractor designed for EEG odor analysis. TensorCSBP is integrated into an explainable feature engineering (XFE) pipeline with four steps: (1) TensorCSBP for feature generation, (2) CWNCA for feature selection, (3) tkNN classifier for decision making, and (4) DLob method for symbolic interpretability. Results: TensorCSBP XFE was evaluated on a newly collected 32-channel EEG dataset for odor detection. It achieved 96.68% accuracy under 10-fold cross-validation. Conclusions: The information entropy of the DLob symbol sequence was 3.5675, demonstrating the richness of the interpretability output. Significance: This study presents a high-accuracy, explainable, and computationally efficient model for EEG-based odor classification. TensorCSBP bridges low-level signal patterns with symbolic neuroscience insights, offering real-time potential for BCI and clinical applications.

Figures