Henrike Stephani

Fraunhofer Institute for Industrial Mathematics

Author Statistics

Papers

Citation

H-Index

i-10 index

Research Trends

Author Order

Document Type

Co-Authors

Janis Keuper

Offenburg University of Applied Sciences

Thomas Weibel

Fraunhofer Institute for Industrial Mathematics

Lars Nieradzik

Fraunhofer Institute for Industrial Mathematics

Stephanie Helmling

Johann Heinrich von Thünen-Institut

Jördis Sieburg-Rockel

Hamburg Institute of International Economics

Bettina Heise

Research Center for Non Destructive Testing (Austria)

Ali Moghiseh

Fraunhofer Institute for Industrial Mathematics

Lars Nieradzik

Lund University

Karin Wiesauer

Ordensklinikum Linz Barmherzige Schwestern

Ronald Rösch

Fraunhofer Institute for Industrial Mathematics

Cooperative Institutions

Fraunhofer Institute for Industrial Mathematics

University of Kaiserslautern

Fraunhofer Society

Fraunhofer Institute for Physical Measurement Techniques

Johannes Kepler University of Linz

Hamburg Institute of International Economics

Research Center for Non Destructive Testing (Austria)

Daimler (Germany)

Rheinland-Pfälzische Technische Universität Kaiserslautern-Landau

University of Applied Sciences Upper Austria

Author Statistics

Papers

Citation

H-Index

i-10 index

Research Field

Wavelet-Based Dimensionality Reduction for Hyperspectral THz Imaging

Henrike Stephani Michael Herrmann Frank Bauer Bettina Heise

With terahertz time-domain spectroscopy, hyperspectral images can be acquired where each pixel contains a full spectrum of the range of several terahertz (THz). An enormous amount of data is generated. Therefore, advanced methods for automated data analysis and image processing are required. We present a wavelet-based approach for channel reduction and feature selection for a subsequent clustering leading to an image segmentation. The main focus of our method is set on the appropriate dimensionality reduction adapted to the THz spectral characteristics of the samples under investigation. A feature reduction to less than 5% is achieved, thereby enabling a channel-wise image processing on the reduced data set. Furthermore, unsupervised classification is chosen for an automatized segmentation including all channel information represented in the wavelet domain. Relevant characteristics of the THz spectra are preserved by our feature selection, in particular the distribution of the peak position and peak depth. The proposed method for channel reduction is verified by extensive simulations at first. Finally, it is demonstrated on various real-world measurements of chemical compounds. The improved performance of the analysis on the reduced feature set could be shown in comparison with the evaluation on the full data set.

Feature (linguistics)

Data set

Cite

Citations (5)

Challenging the Black Box: A Comprehensive Evaluation of Attribution Maps of CNN Applications in Agriculture and Forestry

Proceedings of the 17th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (2024)

Lars Nieradzik Henrike Stephani Jördis Sieburg-Rockel Stephanie Helmling Andrea Olbrich

Black box

10.5220/0012363400003660

Cite

Citations (1)

Top-GAP: Integrating Size Priors in CNNs for more Interpretability, Robustness, and Bias Mitigation

arXiv (Cornell University) (2024)

Lars Nieradzik Henrike Stephani Janis Keuper

This paper introduces Top-GAP, a novel regularization technique that enhances the explainability and robustness of convolutional neural networks. By constraining the spatial size of the learned feature representation, our method forces the network to focus on the most salient image regions, effectively reducing background influence. Using adversarial attacks and the Effective Receptive Field, we show that Top-GAP directs more attention towards object pixels rather than the background. This leads to enhanced interpretability and robustness. We achieve over 50% robust accuracy on CIFAR-10 with PGD $\epsilon=\frac{8}{255}$ and $20$ iterations while maintaining the original clean accuracy. Furthermore, we see increases of up to 5% accuracy against distribution shifts. Our approach also yields more precise object localization, as evidenced by up to 25% improvement in Intersection over Union (IOU) compared to methods like GradCAM and Recipro-CAM.

Interpretability

Robustness

10.48550/arxiv.2409.04819

Cite

Citations (0)

Automatically Detecting Peaks in Terahertz Time-Domain Spectroscopy

Henrike Stephani Joachim Jonuscheit Christoph Robiné Bettina Heise

To classify spectroscopic measurements it is necessary to have comparable methods of evaluation. In Terahertz (THz) time-domain spectroscopy, as a new technology, neither the presentation of the data nor the peak detection is standardized yet. We propose a procedure for automatic peak extraction in THz spectra of chemical compounds. After preprocessing in the time-domain, we use a variance based algorithm for determining the valid frequency region. We furthermore propose a baseline correction using simulated THz spectra. We illustrate how this procedure works on the example of hyperspectral THz measurements of six chemical compounds. Subsequently we propose to use unsupervised classification on the thus processed data to robustly detect the characteristic peaks of a compound.

Terahertz time-domain spectroscopy

10.1109/icpr.2010.1085

Cite

Citations (7)

Machine Learning for Inline Surface Inspection Systems: Challenges, Approaches, and Application Example

Lecture notes in statistics (2024)

Henrike Stephani Thomas Weibel Ronald Rösch Ali Moghiseh

10.1007/978-3-031-66253-9_2

Cite

Citations (0)

Automating Wood Species Detection and Classification in Microscopic Images of Fibrous Materials with Deep Learning

arXiv (Cornell University) (2023)

Lars Nieradzik Jördis Sieburg-Rockel Stephanie Helmling Janis Keuper Thomas Weibel

We have developed a methodology for the systematic generation of a large image dataset of macerated wood references, which we used to generate image data for nine hardwood genera. This is the basis for a substantial approach to automate, for the first time, the identification of hardwood species in microscopic images of fibrous materials by deep learning. Our methodology includes a flexible pipeline for easy annotation of vessel elements. We compare the performance of different neural network architectures and hyperparameters. Our proposed method performs similarly well to human experts. In the future, this will improve controls on global wood fiber product flows to protect forests.

Hyperparameter

Identification

10.48550/arxiv.2307.09588

Cite

Citations (0)

Challenges and Approaches when Realizing Online Surface Inspection Systems with Deep Learning Algorithms

Research Square (Research Square) (2022)

Henrike Stephani Thomas Weibel Ronald Rösch Ali Moghiseh

Abstract Using deep learning in complex online surface inspection systems is challenging due to different framework conditions. First, time restrictions in production are usually fixed in terms of clock rate and response time. Furthermore, these methods need a lot of data, while typically the data situation is thin in the beginning as well as continuously unbalanced: defects occur rarely and thereby providing few example data for learning, while the desired detection rate is 100%. Another important issue is that although defect catalogues exist, they often change, especially when automatic inspection is applied for the first time. This is due to imaging systems usually being able to detect more defects than visual-manual inspection, therefore production, management, and quality assurance usually reiterate their prior defect catalogues. However, data driven methods depend heavily on consistent annotation. Therefore, respective parties must be made aware of this issue on the one hand, on the other hand, annotation and reannotation must be easy and useable by non-experts. Related is the issue of parametrization and traceability. Both are not inherent to neural networks but must be provided to some level to help building trust in machine learning methods. In this paper, we present a quality inspection system that uses deep neural networks for defect detection under real production conditions in wood manufacturing. We will address how we systematically deal with the above issues both in terms of process and algorithm.

Traceability

Visual inspection

10.21203/rs.3.rs-2254738/v1

Cite

Citations (0)

Potentials in the Field of Mobility by Mathematical Methods of AI

Springer eBooks (2023)

Anita Schöbel Henrike Stephani Michael Burger

10.1007/978-3-658-40232-7_26

Cite

Citations (0)

Detecting Healthcare Fraud Using Hybrid Machine Learning for Document Digitization

Lecture notes in statistics (2024)

Yannick Becker Elisabeth Leoff Henrike Stephani Thomas Weibel

Digitization

10.1007/978-3-031-66253-9_5

Cite

Citations (0)

Challenging the Black Box: A Comprehensive Evaluation of Attribution Maps of CNN Applications in Agriculture and Forestry

arXiv (Cornell University) (2024)

Lars Nieradzik Henrike Stephani Jördis Sieburg-Rockel Stephanie Helmling Andrea Olbrich

In this study, we explore the explainability of neural networks in agriculture and forestry, specifically in fertilizer treatment classification and wood identification. The opaque nature of these models, often considered 'black boxes', is addressed through an extensive evaluation of state-of-the-art Attribution Maps (AMs), also known as class activation maps (CAMs) or saliency maps. Our comprehensive qualitative and quantitative analysis of these AMs uncovers critical practical limitations. Findings reveal that AMs frequently fail to consistently highlight crucial features and often misalign with the features considered important by domain experts. These discrepancies raise substantial questions about the utility of AMs in understanding the decision-making process of neural networks. Our study provides critical insights into the trustworthiness and practicality of AMs within the agriculture and forestry sectors, thus facilitating a better understanding of neural networks in these application areas.

Black box

10.48550/arxiv.2402.11670

Cite

Citations (0)