Skip to main content
Fachgebiete
Automobil + Motoren Bauwesen + Immobilien Business IT + Informatik Elektrotechnik + Elektronik Energie + Nachhaltigkeit Finance + Banking Management + Führung Marketing + Vertrieb Maschinenbau + Werkstoffe Versicherung + Risiko
DE
EN
Themenseiten
Organisationspsychologie Projektmanagement Marketing Smart Manufacturing
Bücher
Zeitschriften
Weitere Formate
Webinare Technik Webinare Wirtschaft Kongresse Veranstaltungskalender Awards
Jetzt Einzelzugang starten
Zugang für Unternehmen
Referenzkunden
ATZ-Fachkongress

Chassis.tech plus 2024


4 Kongresse in einer Veranstaltung

Treffen Sie in München die Fachleute von Chassis, Lenkung, Bremsen sowie Reifen/Räder.
Erweiterte Suche
Anmelden
nach oben

2024 | Buch

Simsiam Network Based Self-supervised Model for Sign Language Recognition Kapitel lesen Erstes Kapitel lesen

Intelligent Systems and Pattern Recognition

Third International Conference, ISPR 2023, Hammamet, Tunisia, May 11–13, 2023, Revised Selected Papers, Part II

herausgegeben von: Akram Bennour, Ahmed Bouridane, Lotfi Chaari

Verlag: Springer Nature Switzerland

Buchreihe : Communications in Computer and Information Science

Enthalten in: Springer Professional "Wirtschaft+Technik" , Springer Professional "Technik" , Springer Professional "Wirtschaft"

Einloggen, um Zugang zu erhalten
insite
SUCHEN

Über dieses Buch

This volume constitutes selected papers presented during the Third International Conference on Intelligent Systems and Pattern Recognition, ISPR 2023, held in Hammamet, Tunisia, in May 2023.
The 44 full papers presented were thoroughly reviewed and selected from the 129 submissions. The papers are organized in the following topical sections: computer vision; data mining; pattern recognition; machine and deep learning.

Inhaltsverzeichnis

Frontmatter

Pattern Recognition

Frontmatter
Simsiam Network Based Self-supervised Model for Sign Language Recognition
Abstract
To make a more accurate and robust deep learning model, more labeled data is required. Unfortunately, in many areas, it’s very difficult to manage properly labeled data. Sign language recognition is one of the challenging areas of computer vision, to make a successful deep learning model to recognize sign gestures in real-time, a huge amount of labeled data is needed. Authors have proposed a self-supervised learning approach to address this problem. The proposed architecture used Resnet50 v1 backbone-based simsiam encoder network to learn the similarity between two different images of the same class. Calculated cosine similarity passes to MLP head for further classification. The proposed study uses Indian and American Sign Language detests for simulation. The proposed methodology successfully achieve 74.59% of accuracy. Authors have also demonstrated the impact of other self-supervised deep learning models for sign language recognition.
Deep R. Kothadiya, Chintan M. Bhatt, Imad Rida
Study of Support Set Generation Techniques in LAD for Intrusion Detection
Abstract
Support Set generation is an essential process in the Logical Analysis of Data (LAD). The process of binarization results in an increase in the dimensions of the dataset, which can make the classification process more challenging. The support set generation step is performed to select the important features from the binarized dataset. In this paper, five techniques, namely Set covering problem, Mutual Information Greedy algorithm, Information Gain, Gain ratio, and Gini Index, are used to find the minimal support set for the classification of the Intrusion Detection dataset. LAD uses partially defined Boolean functions to generate positive and negative patterns from the historical observations, which are then transformed into rules for the classification of future observations. The LAD classifier is built using different techniques, and their performances on the NSL-KDD dataset are recorded.
Sneha Chauhan, Sugata Gangopadhyay, Aditi Kar Gangopadhyay
Minimal Window Duration for Identifying Cognitive Decline Using Movement-Related Versus Rest-State EEG
Abstract
Until recently, diagnosing people with neurophysiological disorders such as mild cognitive impairment (MCI) was challenging. The common diagnostic techniques used are invasive in nature and time-consuming due to their reliance on the intervention of an expert neuropsychologist and manual diagnosis. Therefore, the adoption of artificial intelligence (AI) and especially machine learning (ML) has proven most useful. It provided healthcare practitioners with an effective tool to diagnose patients faster with higher accuracy. In this paper, a method to separate MCI subjects from healthy controls using movement-related Raw electroencephalogram (EEG) is evaluated, and a new effective EEG segment length is discovered. A variety of binary classifiers are trained and our proposed segment length of 12 s with 50% overlap produces an accuracy of 97.27%.
Basma Jalloul, Siwar Chaabene, Bassem Bouaziz
Modeling Graphene Extraction Process Using Generative Diffusion Models
Abstract
Graphene, a two-dimensional material composed of carbon atoms arranged in a hexagonal lattice, possess a unique array of properties that make it a highly sought-after material for a wide range of applications. Its extraction process, a chemical reaction’s result, is represented as an image that shows areas of the synthesized material. Knowing the initial conditions (oxidizer) the synthesis result could be modeled by generating possible visual outcomes. A novel text2image pipeline to generate experimental images from chemical oxidizers is proposed. Key components of a such pipeline are a textual input encoder and a conditional generative model. In this work, the capabilities of certain text model and generative diffusion model are investigated and some conclusions are drawn providing further suggestions for further full text2image pipeline development
Modestas Grazys
Bird Species Recognition in Soundscapes with Self-supervised Pre-training
Abstract
Biodiversity monitoring related to bird species is often performed by identifying bird species in soundscapes recorded by microphones placed in the birds’ natural habitats. This typically produces a large amount of unlabeled data. While self-supervised machine learning methods have recently been successfully applied to computer vision and natural language processing tasks, state-of-the-art automatic approaches for bird species recognition in audio recordings mainly rely on transfer learning using pre-trained ImageNet models. In this paper, self-supervised learning is leveraged to improve bird species recognition in soundscapes. Specifically, we use a novel self-supervised approach to pre-train a self-attention neural network architecture on the target domain to take advantage of the vast amount of unlabeled and weakly labeled data. Experiments on data sets from different recording environments show the effectiveness of our approach. In particular, self-supervised pre-training on the target domain improves the cross-domain recognition quality.
Hicham Bellafkir, Markus Vogelbacher, Daniel Schneider, Valeryia Kizik, Markus Mühling, Bernd Freisleben
On the Different Concepts and Taxonomies of eXplainable Artificial Intelligence
Abstract
Presently, Artificial Intelligence (AI) has seen a significant shift in focus towards the design and development of interpretable or explainable intelligent systems. This shift was boosted by the fact that AI and especially the Machine Learning (ML) field models are, currently, more complex to understand due to the large amount of the treated data. However, the interchangeable misuse of XAI concepts mainly “interpretability” and “explainability” was a hindrance to the establishment of common grounds for them. Hence, given the importance of this domain, we present an overview on XAI, in this paper, in which we focus on clarifying its misused concepts. We also present the interpretability levels, some taxonomies of the literature on XAI techniques as well as some recent XAI applications.
Arwa Kochkach, Saoussen Belhadj Kacem, Sabeur Elkosantini, Seongkwan M. Lee, Wonho Suh
Classifying Alzheimer Disease Using Resting State Coefficient of Variance BOLD Signals
Abstract
There is a strong relationship between neurodegenerative disease and altering brain clearance. Brain clearance was shown to be driven by three physiological brain signals. In this paper, we show the importance of abnormality detection of physiological pulsations which is effective in detecting neurodegenerative diseases. We use the coefficient of variance of rs-fMRI BOLD signal (CVBOLD) as the physiological signal distribution that reflects brain activity. In this work, we found relations between the CVBOLD values in specific brain regions of Alzheimer’s subjects. We provided a comparison of different models that were used to handle this challenge and the brain regions that are most affected.
Youssef Hosni, Ahmed Elabasy
Proteus Based Automatic Irrigation System
Abstract
The Traditional agricultural strategies are not satisfactory to cope with food security. This field must benefit from latest technologies.
Automatic irrigation is one of the most promising solution to maintain food security; recently, there is a growing interest in this system around the world. It has become possible to establish self-contained decision-making systems that monitor various phenomena by relying on wireless sensor networks, with the possibility of connecting them to the Internet.
The object behind this work, is to develop an automatic irrigation system based on DHT11 temperature and humidity sensor, and Arduino microcontroller. The system is tested using Proteus simulation software and the obtained results were satisfactory.
Wafa Difallah, Sabira Nour, Abdeldjalil Yaga, Isaac Elgoul
How AI can Advance Model Driven Engineering Method ?
Abstract
Artificial intelligence (AI) skills are being increasingly applied in today’s field of computer science. This aims at better satisfying customer requirements, reducing errors, improving decision-making, tackling complex problems, system automation, increasing operational efficiencies, etc. To do so, AI implies several sub-fields such as Machine Learning (ML), Deep Learning (DL), Neural Networks (NN), Natural Language Processing (NLP), Robotics, etc. Applications of AI are innumerable, including healthcare and biomedicine, bio-informatics, physics, robotics, geo-sciences and more. Our current paper studies AI applications for modeling IoT systems using Model Driven Engineering (MDE) method. We survey the most significant research work related to our topic and investigate how AI techniques could be used to better resolve software engineering issues. In the context of the current paper, we particularly focus on healthcare systems as an illustrative specific domain.
Mohamad Suhairi Md Subhi, Willem Nicolas, Akina Renard, Gabriela Maria Garcia Romero, Meriem Ouederni, Lotfi Chaari

Machine and Deep Learning

Frontmatter
Detection of DoS Attacks in MQTT Environment
Abstract
The Message Queuing Telemetry Transport protocol (MQTT) is one of the most widely used application protocols to facilitate machine-to-machine communication in an IoT environment. MQTT was built upon TCP/IP protocol and requires minimal resources since it is lightweight and efficient which makes it suitable for both domestic and industrial applications. However, the popularity and openness of this protocol make it vulnerable and exposed to a variety of assaults, including Denial of Service (DoS) attacks that can severely affect healthcare or manufacturing services. Thus, securing MQTT-based systems needs to develop a novel, effective, and adaptive intrusion detection approach. In this paper, we focus on MQTT’s flaws that allow hackers to take control of MQTT devices. We investigate a deep learning-based intrusion detection system to identify malicious behavior during communication between IoT devices, using an open source dataset namely ‘MQTT dataset’. The findings demonstrate that the suggested approach is more accurate than traditional approaches based on machine learning; with an accuracy rate greater than 99% and an F1-score greater than 98%.
Hayette Zeghida, Mehdi Boulaiche, Ramdane Chikh
Staged Reinforcement Learning for Complex Tasks Through Decomposed Environments
Abstract
Reinforcement Learning (RL) is an area of growing interest in the field of artificial intelligence due to its many notable applications in diverse fields. Particularly within the context of intelligent vehicle control, RL has made impressive progress. However, currently it is still in simulated controlled environments where RL can achieve its full super-human potential. Although how to apply simulation experience in real scenarios has been studied, how to approximate simulated problems to the real dynamic problems is still a challenge. In this paper, we discuss two methods that approximate RL problems to real problems. In the context of traffic junction simulations, we demonstrate that, if we can decompose a complex task into multiple sub-tasks, solving these tasks first can be advantageous to help minimising possible occurrences of catastrophic events in the complex task. From a multi-agent perspective, we introduce a training structuring mechanism that exploits the use of experience learned under the popular paradigm called Centralised Training Decentralised Execution (CTDE). This experience can then be leveraged in fully decentralised settings that are conceptually closer to real settings, where agents often do not have access to a central oracle and must be treated as isolated independent units. The results show that the proposed approaches improve agents performance in complex tasks related to traffic junctions, minimizing potential safety-critical problems that might happen in these scenarios. Although still in simulation, the investigated situations are conceptually closer to real scenarios and thus, with these results, we intend to motivate further research in the subject.
Rafael Pina, Corentin Artaud, Xiaolan Liu, Varuna De Silva
Policy Generation from Latent Embeddings for Reinforcement Learning
Abstract
The human brain endows us with extraordinary capabilities that enable us to create, imagine, and generate anything we desire. Specifically, we have fascinating imaginative skills allowing us to generate fundamental knowledge from abstract concepts. Motivated by these traits, numerous areas of machine learning, notably unsupervised learning and reinforcement learning, have started using such ideas at their core. Nevertheless, these methods do not come without fault. A fundamental issue with reinforcement learning especially now when used with neural networks as function approximators is their limited achievable optimality compared to its uses from tabula rasa. Due to the nature of learning with neural networks, the behaviours achievable for each task are inconsistent and providing a unified approach that enables such optimal policies to exist within a parameter space would facilitate both the learning procedure and the behaviour outcomes. Consequently, we are interested in discovering whether reinforcement learning can be facilitated with unsupervised learning methods in a manner to alleviate this downfall. This work aims to provide an analysis of the feasibility of using generative models to extract learnt reinforcement learning policies (i.e. model parameters) with the intention of conditionally sampling the learnt policy-latent space to generate new policies. We demonstrate that under the current proposed architecture, these models are able to recreate policies on simple tasks whereas fail on more complex ones. We therefore provide a critical analysis of these failures and discuss further improvements which would aid the proliferation of this work.
Corentin Artaud, Rafael Pina, Xiyu Shi, Varuna De-Silva
Deep Learning Models for Aspect-Based Sentiment Analysis Task: A Survey Paper
Abstract
Due to the significant increase in the volume of data shared on the web, Aspect-Based Sentiment Analysis (ABSA) has become essential. This task ensures a detailed sentiment analysis. It identifies firstly the aspect terms (e.g., price, food, etc.) and then classifies their sentiment polarity as positive, negative, or neutral. Many approaches have been used to treat this task including the machine learning-based approach, the rule-based approach, etc. However, with the important increase in the content of the internet, these approaches became relatively unable to analyze this volume of information, resulting in the emergence of the deep learning-based approach which is the subfield of the machine learning-based approach.
Recently many researchers used the deep learning-based approach to address the ABSA. This paper provides a summary of the deep learning models that have been developed for ABSA, as well as a survey of studies that have employed these models to address different subtasks of the ABSA task. Finally, we discuss the implications of our work and potential avenues for future research.
Sarsabene Hammi, Souha Mezghani Hammami, Lamia Hadrich Belguith
A Real-Time Deep UAV Detection Framework Based on a YOLOv8 Perception Module
Abstract
Unmanned Aerial Vehicles (UAVs) or drones are currently gaining a lot of popularity due to the versatility of this technology and its ability to perform multiple tasks in various industries. However, arbitrary or malicious use of drones can pose a major risk for public and aviation safety. The automated detection and neutralization of malicious drones to avoid fatal incidents is therefore of primary interest for aerial security systems. Recently, deep learning based approaches for object detection have gained great attention due to their high prediction accuracy. In the proposed work, we use a deep learning object detection model based on latest versions of YOLO, i.e. v7 and v8, to detect and track one drone by another drone in a pursuit-evasion scenario. The detection accuracy achieved with the YOLOv8 model is 99% Average Precision (AP) and the inference is 107.5 FPS, proving the effectiveness of the proposed approach for real-time UAV detection. A comparative study shows that the YOLOv7 model achieves the same accuracy but with slower inference.
Wided Souid Miled, Moulay A. Akhloufi, Hana Ben Asker

Open Access

A Deep Neural Architecture Search Net-Based Wearable Object Classification System for the Visually Impaired
Abstract
The World Health Organization estimates that a staggering 2.2 billion individuals worldwide suffer from vision impairments, drastically limiting independence and quality of daily life and leading to billions of dollars in direct costs and annual productivity losses. Although the field of machine learning has made significant strides in recent years, particularly in image classification, these advances have predominantly focused on tasks that are visual in nature, which can be challenging for vision-impacted individuals. Much work has been published on obstacle avoidance and large-object detection for the visually impaired. However, little has been done to aid them in better understanding complex indoor daily-living environments. For these reasons, this study develops and presents a wearable object classification system specifically designed to assist the visually impaired in identifying small tabletop objects commonly found in their surrounding indoor environments. Through transfer learning, the system uses a pretrained neural architecture search network called NASNet-Mobile and a custom image dataset to conduct highly effective small-object classification with model accuracies of over 90.00%. The proposed transfer-learning model is subsequently deployed on a wearable wrist device for real-world applicability. This study ultimately evaluates and demonstrates the system’s ability to accurately classify small tabletop objects using an eight-trial experiment that calculates the system’s average precision, recall, and F1 score to be 99.30%, 97.93%, and 98.61%, respectively. Overall, this system represents a significant step forward in the development of machine learning systems that constructively assist the visually impaired while simultaneously improving their daily independence and quality of life.
Aniketh Arvind
PDF Zum Volltext
Multicarrier Waveforms Classification with LDA and CNN for 5G
Abstract
Accurate classification of multi-carrier waveforms is significant for ensuring quality signal reception, improved system throughput, and reduce the power consumption in future wireless generations as 5G. The aim of this paper is to improve the precision classification of various multicarrier waveforms. Here, we propose a novel representation of multicarrier signals in AWGN environment and use suitable networks for classification, which utilizes deep convolutional neural networks to classify OFDM-QAM, and FBMC-OQAM. LDA-based (Linear Discriminant Analysis) method is proposed in this paper to reduce the input dimensions of CNN. The results reveal that LDA-CNN is a promising candidate for wireless communication.
M’hamed Bilal Abidine
Weeds Detection Using Mask R-CNN and Yolov5
Abstract
Agriculture have seen a decrease in yields crops due to weeds, so they were encouraged to use pesticides and chemical treatments to eliminate them and obtain the results despite their environmental damage, another hand the new IA technology can be give best solution help farmers to detect and elimination weeds. The detection of objects is crucial in many computer vision applications, and our focus is on detecting weeds, which pose a significant challenge in agriculture due to their negative impact on harvest yield and quality. In this study, we evaluate two state-of-the-art convolutional neural network-based object detectors for weed detection under real-world conditions without staging. Our evaluation considers both speed and accuracy metrics using a dataset of images captured using a smartphone camera, the performance of weed detection is evaluated according to a predefined geographic location in Tebessa, an Algerian eastern region. Additionally, we compare the performance of these models with and without additional training using examples from different databases.
Merzoug Soltane, Mohamed Ridda Laouar
Boruta-AttLSTM: A Novel Deep Learning Architecture for Soil Moisture Prediction
Abstract
Water scarcity is worsening due to poor water management in irrigated areas, which directly impacts global food safety. Furthermore, effective irrigation scheduling necessitates predicting future soil moisture content, representing soil water availability. For this purpose, the current study proposes a novel data-driven architecture based on deep learning algorithms to predict soil volumetric water content. The proposed architecture combines the time-processing ability of Long Short-Term Memory with the attention mechanism’s ability to process long sequences. The suggested architecture’s resulting model is compared to a 2-layer LSTM in terms of MSE, MAE, RMSE, and R2 score. This study also examines the relationships between various climate and soil parameters and targets soil moisture. The relevance of input features is considered by the feature selection strategy using their computed shapley values. The findings of this study suggest that attention mechanisms can increase the performance and generalizability of regular LSTMs.
Bamory Ahmed Toru Koné, Bassem Bouaziz, Rima Grati, Khouloud Boukadi
Graph Autoencoder with Community Neighborhood Network
Abstract
Neighborhood information can be extracted from graph data structure. The neighborhood is valuable because similar objects tend to be connected. Graph neural networks (GNN) represent the neighborhood in layers depending on their proximity. Graph autoencoders (GAE) learn the lower dimensional representation of graph and reconstruct it afterward. The performance of the GAE might be enhanced with the behavior of GNNs. However, utilizing the neighborhood information is challenging. Far neighbors are capable of building redundantly complex networks due to their decreasing similarity. Yet, less neighborhood models are closer to GAE. Restricting the neighborhood within the same community enriches the GNN. In this work, we propose a new unsupervised model that combines GNN and GAE to improve the representation learning of graphs. We examine the outcomes of the model under different neighborhood configurations and hyperparameters. We also prove that the model is applicable to varying sizes and types of graphs within different categories on both synthetic and published datasets. The outcome of the community neighborhood network is resistant to overfitting with fewer learnable parameters.
Ahmet Tüzen, Yusuf Yaslan
Question-Aware Deep Learning Model for Arabic Machine Reading Comprehension
Abstract
Machine reading comprehension is one of the most long-standing challenges in artificial intelligence. It is a subtask question-answering system that aims to find a text span in a reading context that answers a question. It is the core of several applications, such as customer service systems and chatbots. However, However, few studies have targeted machine reading comprehension of Arabic text. Besides, The existing models generate a context representation independently from the question. As an analogy to some people’s comprehension styles who believe that by reading the question, they gain a better understanding of the reading passage, we present a question-aware neural machine reading comprehension model for Arabic. Our model extracts a representation of the context by incorporating the question using several bidirectional attention units to achieve various levels of question-centered context understanding. Our experimental evaluation shows that our model outperforms strong machine reading comprehension baselines, including DrQA and QaNET models applied to Arabic, by a significant margin with an F1 score of 59.6%.
Marwa Al-Harbi, Rasha Obeidat, Mahmoud Al-Ayyoub, Luay Alawneh
A Hybrid Deep Learning Scheme for Intrusion Detection in the Internet of Things
Abstract
The Internet of Things (IoT) is the connection of smart devices and objects to the internet, allowing them to share and analyze data, communicate with each other, and be controlled remotely. Several IoT devices are designed to collect, process, and store confidential data in order to perform their intended function. This information can be sensitive such as location, health, military, financial information, and biometric data. The efficient implementation of IoT networks has become increasingly reliant on security. In IoT networks, several researchers used intrusion detection systems (IDS) for the identification of cyberattacks where machine learning (ML) and deep learning (DL) are significant components. The existing IDS still needs improvements for the detection of multiclass detection to identify each category of attack separately. To improve the detection performance of IDS, this study proposes a hybrid scheme of convolutional neural networks (CNN) and gated recurrent units (GRU). The proposed hybrid scheme integrates two CNN layers and three GRU layers. The proposed scheme was assessed using the IoTID20 dataset.
Asadullah Momand, Sana Ullah Jan, Naeem Ramzan
Identifying Discourse Markers in French Spoken Corpora: Using Machine Learning and Rule-Based Approaches
Abstract
The objective of this work is to study the identification of French discourse markers (DM), in particular the polyfunctional occurrences such as ‘attetion’, bon, quoi, la preuve. A number of words identified as DM, and traditionally considered as adverbs or interjections, are also, for instance, adjectives or nouns. For example bon can be a DM or an adjective, ‘attetion’ can be a DM or a noun, etc. Hand annotation is in general robust but time consuming. The main difficulty with automatic identification is to take the context of the DM candidate correctly into account. To do that, a mechanisms based on rule-based and machine learning approaches was built, in order to reach an acceptable level of performance and reduce the expert effort. This study will provide a comprehensive use case of a machine learning algorithm, which has proved a good efficiency in dealing with such linguistic phenomena. In addition, an evaluation was done for the Unitex platform in order to determine the efficiency and drawbacks of this platform when dealing with such type of tasks.
Abdelhalim Hafedh Dahou
A Comparative Study of the Impact of Different First Order Optimizers on the Learning Process of UNet for Change Detection Task
Abstract
UNet is an encoder-decoder neural network that has been used to detect changes in remote-sensing images. This paper provides a comparative study on the performance of UNet when trained with different optimizers for the Change Detection task. Although several previous works aim to compare different UNet models for change detection, this paper is, as far as we know, the first work that investigates UNet regarding the optimization method that is used in the learning process. This can help designing of more efficient UNet models, especially with limited training resources. We compare five common gradient-based optimization techniques: Gradient descent with Momentum (Momentum GD), Nesterov Accelerated Gradient (NAG), Adaptive Gradient (AdaGrad), Root Mean Square Propagation optimizer (RMSProp), and the adaptive moment estimation optimizer (Adam). For this purpose, UNet is trained over 200 epochs using ONERA dataset for the optimization of the binary cross entropy. The model is assessed using three metrics: Accuracy, Precision, and F1-score. According to the obtained results, RMSProp, NAG, and AdaGrad reached the highest validation accuracies: 0.976, 0.978, and 0.979 with \(10^{-2}, 10^{-3}\) and \(10^{-4}\) respectively. Adam was the fastest to converge and scored the lowest validation loss. Moreover, Adam scored the highest precision and F1-score across all learning rates, with 0.491 and 0.376 respectively. We also note that both Momentum-based algorithms and adaptive algorithms perform better with relatively small learning rate values.
Basma Dokkar, Bouthaina Meddour, Khadra Bouanane, Mebarka Allaoui, Mohamed Lamine Kherfi
Backmatter
Metadaten
Titel
Intelligent Systems and Pattern Recognition
herausgegeben von
Akram Bennour
Ahmed Bouridane
Lotfi Chaari
Copyright-Jahr
2024
Electronic ISBN
978-3-031-46338-9
Print ISBN
978-3-031-46337-2
DOI
https://doi.org/10.1007/978-3-031-46338-9

Premium Partner

    Bildnachweise