275003 research outputs found
Sort by
Statistical Feature-Based Detection of Adversarial Noise and Patch Attacks in Image and Deepfake Analysis
No full text195208Adversarial attacks pose a significant threat to the reliability and trustworthiness of machine learning systems, particularly in image classification tasks like deepfake detection. This chapter presents a comprehensive approach to detecting two prominent types of adversarial attacks: noise perturbation attacks and patch-based attacks. Using the ImageNet classification task as a primary use case, we investigate methods based on statistical features for identifying adversarial noise across a diverse range of attacks. These methods are designed to detect subtle changes in image distributions caused by adversarial manipulations, offering a lightweight and interpretable solution for adversarial attack detection that can be part of a multi-class detector framework. Building upon this foundation, the chapter explores the security-critical application of deepfake detection. Here, patch-based attacks are examined in depth. The proposed detection framework leverages statistical and spatial features to identify patch artifacts, ensuring robustness against these localized yet highly effective attacks. Our analysis compares the effectiveness of statistical detectors across multiple adversarial attack types and evaluates their performance in real-world scenarios. By addressing both noise perturbation and patch attacks, this chapter provides actionable insights and tools for enhancing the security of machine learning systems deployed in high-stakes applications, bridging the gap between theory and practice in adversarial defense
AI Video Monitoring of Aquatic Organisms - Opportunities, Limits and Best Practice Recommendations
No full text5461Die manuelle Verarbeitung von Fischvideos von Unterwasserkameras erfordert einen erheblichen Personalaufwand, was den Einsatz von videobasiertem Monitoring bisher einschränkt. Fortschritte in der Computer Vision ermöglichen es, mit Methoden der künstlichen Intelligenz (KI) die Umweltbedingungen automatisch zu klassifizieren, das Vorhandensein von Fischen zu erkennen, die Arten zu klassifizieren sowie ihre Gesamtkörperlänge und Schwimmrichtung zu schätzen. Trotz des Versprechens der KI, diese arbeitsintensiven Aufgaben zu automatisieren, gibt es auch erhebliche technische und praktische Herausforderungen. In diesem Artikel stellen wir wichtige Erkenntnisse, Leitlinien und Grenzen für den Einsatz von KI für das automatisierte Monitoring von Fischen mithilfe von Videomonitoring vor.Manual processing of fish videos from underwater cameras can be very time-consuming, which currently limits the use of video-based monitoring. Recent advances in computer vision allow artificial intelligence (AI) methods to automatically classify environmental conditions, detect the presence of fish, classify species and estimate their total body length and swimming direction. However, despite the promises of AI to automate these labor-intensive tasks, there remain important technical and practical challenges. In this article, we present key findings, guidelines and limitations for the use of AI for automated monitoring of river fish using underwater camera videos.1151
Electronic network simulation of 5-junction solar cells and impacts on the ideal current matching
No full textWe combined optical ray tracing simulations of the C3PV module with electronic network simulations of the 5C46 5 junction solar cell to generate IV curves of a single cell under realistic module conditions. These take the effect of the different current generation distributions within the 5 sub cells into account. The simulations were evaluated for different air masses to estimate and compare the daily energy yield in a good CPV location. Subsequently we varied the current matching in the 5 junction solar cell starting with the fixed current balance of the 5C46 solar cell and introducing variation shapes. The best variation proved to be a ramp shaped variation with 4% more current in the first junction J1 and 4% less current in J4. Yet the original 5C46 current balance already proved to be very good - especially when a homogeneous flux on the solar cell is considered
On the Benefit of Dual-Domain Denoising in a Self-Supervised Low-Dose CT Setting
No full textComputed tomography (CT) is routinely used for three-dimensional non-invasive imaging. Numerous data-driven image denoising algorithms were proposed to restore image quality in low-dose acquisitions. However, considerably less research investigates methods already intervening in the raw detector data due to limited access to suitable projection data or correct reconstruction algorithms. In this work, we present an end-to-end trainable CT reconstruction pipeline that contains denoising operators in both the projection and the image domain and that are optimized simultaneously without requiring ground-truth high-dose CT data. Our experiments demonstrate that including an additional projection denoising operator improved the overall denoising performance by 82.4-94.1 %/12.5-41.7 % (PSNR/SSIM) on abdomen CT and 1.5-2.9 %/0.4-0.5 % (PSNR/SSIM) on XRM data relative to the low-dose baseline. We make our entire helical CT reconstruction framework publicly available that contains a raw projection rebinning step to render helical projection data suitable for differentiable fan-beam reconstruction operators and end-to-end learning
Enhancing Transparency in Low-Voltage Grids through ANN-Based Evaluation of Measurement Locations
No full text176180As part of the ongoing energy transition, smart-metering technology will be installed at the low-voltage level. In addition to the deployment of smart meters, MV/LV transformers, feeders in local substations (SS), or feeders of cable distribution cabinets (CDC) will be equipped with measuring devices. However, more technical approaches are needed to evaluate the decision-making process for the placements of measurements. In this study, optimal locations for measurement devices at low-voltage grids are determined using artificial neural network (ANN) estimations. Time series simulations are computed using secondary data to provide training and test sets. The trained ANNs determine the quality of each measuring location based on estimation errors. The results of the analyses demonstrate that the methodology can support a focused deployment of measurement devices and thus contribute to an increase in grid transparency. Furthermore, measurements must be positioned individually for each LV grid, as the estimation results significantly depend on the underlying secondary data
Structural and ferroelectric properties of Al1−xScxN
No full text7199The wurtzite-type solid solutions can be considered the newest class of thin film ferroelectrics available to microelectronics. They are characterized by very high spontaneous polarizations in excess of 100 μC/cm2, excellent retention due to their high coercive fields and straight forward compatibility to silicon and GaN based semiconductor platforms. Their far from common scalability in terms of film thickness over more than 3 orders of magnitude—from below 5 nm to above 500 nm, further make then suitable for a wide variety of applications, from memory to micromechanical sensors and actuators. In this chapter, we give the reader an introduction to the first, and as of now predominant, wurtzite-type ferroelectric: Al1−xScxN. We start by discussing the effect of the chemical composition of the solid solution on the structure of the compound and how the structure simultaneously is also dependent on the deposition process and the underlying growth template. This is followed by a summary of the ferroelectric properties in dependence of Scandium content and stress, their endurance and retention up to temperatures above 1000 °C, while also discussing the main drawback which currently still affects the wurtzite-type ferroelectrics: Their comparable high leakage current in the vicinity and above the coercive field. We conclude by comparing the occurrence of ferroelectricity in Al1−xScxN with other recently confirmed wurtzite-type ferroelectrics.11
Process Dependent Material Characterization for Warpage Control of Fan-Out Wafer Level Packaging
No full text21652170Fan-out Wafer-Level Packaging (FO-WLP) is seen as an enabling technology for new packaging options in heterogeneous integration. For the RDL-last approach the chips are embedded in a mold process on a carrier and the electrical connection is realized with redistribution layers (RDL) in a wafer level process, which enables high precision interconnects. A major challenge within this process is to control the warpage of the molded wafer. As in future a multitude of different package types and applications are being addressed, this warpage control needs to be realized for different materials, process conditions and package designs. Therefore, a precise simulation technology becomes necessary to assess the influence of different design features on warpage. This paper focusses on the analysis of warpage behavior throughout the first crucial process steps. An example mold material is comprehensively analyzed in order to identify crucial steps in the planned modelling approach. The analysis covers the temperatures and times applied in post mold cure (150°C), thermal release (200°C) and RDL cure (250°C) with a special focus on 200°C. The warpage is measured on a sample combining mold and a silicon wafer with a diameter of 200 mm. Throughout these steps, the development of warpage is quantified and it is shown that an irreversible warpage behavior (hysteresis) is highly relevant. With the combination of thermomechanical measurements (TMA, DMA) and finite element modelling the origin of this effect is being identified. Thus a new approach is suggested to characterize and model warpage in processing of molded wafers at elevated temperatures and a basis is given to control warpage and improve the prediction capability of finite element models
Deposition and properties of sputtered phosphorus doped amorphous silicon films for passivating contacts
No full textThe preparation of phosphorus doped silicon films by DC sputtering in an Ar/PH3 atmosphere was investigated. The deposition process is determined by the dissociation of the PH3 molecule. The deposited films have active dopant concentrations >1 1020 cm-3 after annealing. However, the dissociation also produces hydrogen. The hydrogen trapped in the film segregates during annealing and causes blistering. The degree of blistering can be minimized by adjusting process parameters and sequence as well as film thickness. A maximum iVoc of 741 mV after annealing and hydrogen passivation is measured on symmetric tunnel oxide plus poly-silicon lifetime structures processed on n-type FZ silicon wafers
Integration of Photo-Imaging Technology and Microvias in LTCC for Enhanced High-Frequency Applications and Packaging
No full textIn low-temperature co-fired ceramic (LTCC) technology, there is a growing demand for higher resolution of printed structures, particularly below 50μm, to meet the miniaturization needs of high-frequency applications such as 5G, 6G, or radars. To address these challenges, Fraunhofer IKTS employs photo-imaging (PI) technology, achieving resolutions down to 15μm in the cofiring process. In addition to planar structuring, the miniaturization of vias is essential, as standard pastes are insufficient for filling diameters below 100μm. This work aims to establish a robust micro-multilayer technology process chain that combines PI-based fine-line metallization with microvias (30-50μm) in LTCC. Two approaches for via formation were investigated: picosecond laser drilling and the use of dielectric PI pastes. The laser direct imaging (LDI) process demonstrated high structural quality and enabled spatially resolved shrinkage compensation during sintering. This allows for precise realization of microstructures ensuring the geometric accuracy required for high-frequency circuit applications
SOI Integrated Micromagnets for Mechanical Magnetic Field Detection
No full textThe integration of micromagnets into micromechanical systems presents promising opportunities for the development of passive magnetic field sensors that do not rely on continuous electrical energy. Therefore, micromagnetic transducers are required, which allow a conversion of external magnetic fields into a force or displacement that can be further processed within a micromechanical system. In this letter, we present the integration of powder-based permanent magnets in silicon-on-insulator technology on wafer-level and with in-plane magnetization. Based on this technology, demonstrators consisting of a magnetic actuator and a mechanical amplifier are designed, fabricated, and characterized. Measurement results show that external magnetic fields of about 223 mT enable either the generation of a displacement of 76 μm at the mechanical amplifier or a force of 1.68 mN.7