Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Empirical Analysis of Anomaly Detection on Hyperspectral Imaging Using Dimension Reduction Methods

Dongeon Kim, YeongHyeon Park

TL;DR

Addresses latency and explainability challenges in hyperspectral imaging (HSI) for anomaly detection in manufacturing. Proposes a feature-selection approach (FI/PI) as an alternative to feature extraction methods like PCA, and evaluates on synthetic HSI derived from the MVTec AD dataset. Finds that FI/PI-based channel selection achieves comparable or better anomaly detection performance while delivering up to 6.90× faster inference compared to full-spectrum HSI. This work supports cost-efficient spectroscopic camera design by identifying the most informative wavelengths and preserving detection accuracy.

Abstract

Recent studies try to use hyperspectral imaging (HSI) to detect foreign matters in products because it enables to visualize the invisible wavelengths including ultraviolet and infrared. Considering the enormous image channels of the HSI, several dimension reduction methods-e.g., PCA or UMAP-can be considered to reduce but those cannot ease the fundamental limitations, as follows: (1) latency of HSI capturing. (2) less explanation ability of the important channels. In this paper, to circumvent the aforementioned methods, one of the ways to channel reduction, on anomaly detection proposed HSI. Different from feature extraction methods (i.e., PCA or UMAP), feature selection can sort the feature by impact and show better explainability so we might redesign the task-optimized and cost-effective spectroscopic camera. Via the extensive experiment results with synthesized MVTec AD dataset, we confirm that the feature selection method shows 6.90x faster at the inference phase compared with feature extraction-based approaches while preserving anomaly detection performance. Ultimately, we conclude the advantage of feature selection which is effective yet fast.

Empirical Analysis of Anomaly Detection on Hyperspectral Imaging Using Dimension Reduction Methods

TL;DR

Addresses latency and explainability challenges in hyperspectral imaging (HSI) for anomaly detection in manufacturing. Proposes a feature-selection approach (FI/PI) as an alternative to feature extraction methods like PCA, and evaluates on synthetic HSI derived from the MVTec AD dataset. Finds that FI/PI-based channel selection achieves comparable or better anomaly detection performance while delivering up to 6.90× faster inference compared to full-spectrum HSI. This work supports cost-efficient spectroscopic camera design by identifying the most informative wavelengths and preserving detection accuracy.

Abstract

Recent studies try to use hyperspectral imaging (HSI) to detect foreign matters in products because it enables to visualize the invisible wavelengths including ultraviolet and infrared. Considering the enormous image channels of the HSI, several dimension reduction methods-e.g., PCA or UMAP-can be considered to reduce but those cannot ease the fundamental limitations, as follows: (1) latency of HSI capturing. (2) less explanation ability of the important channels. In this paper, to circumvent the aforementioned methods, one of the ways to channel reduction, on anomaly detection proposed HSI. Different from feature extraction methods (i.e., PCA or UMAP), feature selection can sort the feature by impact and show better explainability so we might redesign the task-optimized and cost-effective spectroscopic camera. Via the extensive experiment results with synthesized MVTec AD dataset, we confirm that the feature selection method shows 6.90x faster at the inference phase compared with feature extraction-based approaches while preserving anomaly detection performance. Ultimately, we conclude the advantage of feature selection which is effective yet fast.
Paper Structure (10 sections, 4 equations, 4 figures, 3 tables, 1 algorithm)

This paper contains 10 sections, 4 equations, 4 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Method
  3. Preprocessing
  4. Anomaly detection
  5. Results
  6. Dataset and evaluation metrics
  7. Experiment settings
  8. Comparison of model performance
  9. Latency analysis
  10. Conclusion

Figures (4)

  • Figure 1: An illustration of a schematic framework for comparison of anomaly detection performance with dimension reduction methods.
  • Figure 2: Each (a) and (b) show the sample spectrum of original RGB image and synthesized HSI, respectively. Original spectrum only can represent three wavelengths. The spectrum of HSI that synthesized by interpolation is consisted with 300 wavelengths.
  • Figure 3: Single channel images extracted from synthesized HSI. (a), (b), (c) indicate the orange (660.98nm), red (751.85nm), and near-infrared ranges (1015.52nm), respectively. Since the capability to identify the anomalous region is different for each channel, we should selectively use the important channels.
  • Figure 4: A result of measured feature importance for carpet dataset. Within all channels, we discard all channels except top six important channels for reducing channel dimension.