Supervised classification methods rely heavily on labeled training data. However, errors in the manually labeled data arise inevitably in practice, especially in applications where data labeling is a complex and expensive process, as is often the case in remote sensing. Erroneous labels affect the learning models, deteriorate the classification performances and hinder thereby subsequent image analysis and scene interpretation. In this paper, we analyze the effect of erroneous labels on spectral signatures of landcover classes in remotely sensed hyperspectral images (HSIs). We analyze also statistical distributions of the principal components of HSIs under label noise in order to interpret the deterioration of the classification performance. We compare the behaviour of different types of classifiers: spectral only and spectral-spatial classifiers based on different learning models including deep learning. Our analysis reveals which levels of label noise are acceptable for a given tolerance in the classification accuracy and how robust are different learning models in this respect.
The Internet of Things is changing the approach to data transmission and protocol design as well as network services. That is why it is considered to be another technological revolution. The challenge faced by designers of IoT solutions is to determine the scalability of a given technology, with particular emphasis on unlicensed bandwidth (ISM) transmission in highly urbanized areas. Because the construction and implementation of a wireless network for Internet of Things in each of the presented technologies is expensive and time consuming, it must be preceded by an assessment of performance using computer simulations. The literature contains various approaches to modeling the mechanisms of the MAC layer of LoRa technology and implementation in the LoRaWAN network. The article provides an overview of representative LoRa MAC network simulators. It presents and comments the most important research results obtained by the authors of the mentioned simulators.
This paper considers a system of macro diversity (MD) that consists of a macro receiver of SC diversity (MD SC) and two micro MRC (mD MRC) receivers, which operate in a correlated Gamma-shadowed Rayleigh multipath fading environment. The combination of the maximum L-branch ratio (SNR signal-to-noise) is realized at the micro level, and the optional selection combiner (SC) with two base stations is performed at the macro level. The closed-form expression for the moment-generating function (MGF) of the SC macro diversity output signal envelope is calculated. This result is used to study important system performance parameters such as downtime probability, average bit error probability (ABEP), average output signal value, and amount of fading (AoF). The paper also calculates the channel capacity (CC) of the MD SC receiver. The results are graphically illustrated showing the influence of di ff erent system parameters on performance, as well as the improvement due to the use of a combination of micro and macro diversity systems. Furthermore, the derived expressions are leveraged within the GPU-enabled mobile network modeling, planning and simulation environment for Quality of Service (QoS) parameter value determination.
Deep learning techniques are currently gaining high prominence in the field of computer music generation. The main objective of this paper is to propose a novel, imagebased, data representation for music that is tailored for training deep learning models. Specifically, we suggest using color encodings to represent music notes inside images. We develop an intelligent accompaniment music generator using the pix2pix Generative Adversarial Network (GAN). We compare the effect of our suggested data representation technique on the pix2pix network learning as opposed to the traditional binary encoding scheme used in the literature. We also suggest a post-processing technique for enhancing the quality of the generated music. Additionally, we introduce two automatic evaluation metrics for assessing the generated music based on dissimilarity and musical harmony. Finally, we suggest a variation of our proposed data representation and devise a comparison between the two representations. Our experimental results show that our music representation achieved better results on pix2pix GANs over the traditional representations, reaching a loss function value of 0.001. Moreover, by evaluating the music generated by our system, the results show that our proposed post-processing technique enhanced the musical harmony and the dissimilarity of the generated music by 51.26% and 81.98% respectively.
The Internet of Things systems, as all kinds of networks, are susceptible to various kinds of attacks on 5G network, which hinder their functionalities and pose a threat to the security of their users. One of such attacks are Distributed Denial of Service Attacks, which are able to block the whole network and disable the functions of the devices within it. The method of recognizing and neutralizing such attacks presented in this paper is based on Ordered Fuzzy Numbers and it is easy for implementation.
In the near future, the demand of wireless communications will elevate tremendously. The rapid increasing number of mobile end devices will require a much higher data rate connection than nowadays e.g. to smart homes (Internet of Things, IoT) or to the Internet. The radiation power pattern of base stations and mobile end devices will completely change for the 5G Next Generation Mobile Network technology, which is expected to use frequency bands up to 100 GHz. The electro-magnetic exposure especially to human bodies will increase in the future, because most of the wireless Internet connections are realized in RF technology. In particular, the technology standard 5G and others are being promoted and realised worldwide, which rises a public discussion regarding the relevant non-ionizing electromagnetic radiation exposure for the population. To compare the everyday exposure situation to di ff erent RF radiation sources six di ff erent measurement campaigns are presented in this contribution. In particular this publication presents a comparison of the non-ionizing electromagnetic radiation exposure between a mobile base station, a mobile phone, mobile phone radiation in a moving vehicle, WLAN base stations, state of the art TV broadcast transmitters like DVB-T2 and 5G base stations as part of long-term measurements. In particular, the di ff erent power flux densities of the technologies mentioned are measured, compared and discussed with regard to the legal framework and limits.
In this paper, we propose an e ffi cient method for learning a local image descriptor and its inversion function using a modified version of a variational autoencoder (VAE) - a β -VAE. We examine di ff erent values of β in the loss function of the β -VAE to find the an optimal balance between incentivising the similarities between input patches to be preserved in latent space, and ensuring good reconstruction of the patches from their encodings in latent space. Our proposed descriptor demonstrates patch retrieval comparable to the reference autoencoder-based local image descriptor, and also shows improved reconstruction of patches from their encodings.
Video based heart rate estimation based on the PPG technique is a remote optical technique which allows us to determine the heart and breath rates through the intensity of motion variations. This paper proposes a new monitoring method for estimating the heart rate using a vision system with the VIS and SWIR camera.
Local image descriptors play a crucial role in many image processing tasks, such as object tracking, object recognition, panorama stitching, and image retrieval. In this paper, we focus on learning local image descriptors in an unsupervised way, using autoencoders and variational autoencoders. We perform a thorough comparative analysis of these two approaches along with an in-depth analysis of the most relevant hyperparameters to guide their optimal selection. In addition to this analysis, we give insights into the difficulties and the importance of selecting right evaluation techniques during the unsupervised learning of the local image descriptors. We explore the extent to which a simple perceptual metric during training can predict the performance on tasks such as patch matching, retrieval and verification. Finally, we propose an improvement to the encoder architecture that yields significant savings in memory complexity, especially in single-image tasks. As a proof of concept, we integrate our descriptor into an inpainting algorithm and illustrate its results when applied to the virtual restoration of master paintings. The source code required to reproduce the presented results has been made available as a repository on GitHub (https://github.com/nimpy/local-img-descr-ae).
This paper presents current possibilities of using stereophotogrammetry to scan interiors for video games. This technology offers great possibilities for effective room scanning based on popular cameras, however, it also has some limitations in the implementation of a fully automatic image processing. As the result of the STERIO project, the Sterio Reconstruction tool has been created that offers semi-automatic 3D reconstruction of building interiors. The proposed technology can be used, among others, for modeling rooms for video games which action takes place in locations reflecting real interiors. The original, large-scale experiment has been carried out in order to validate the created technology in real conditions. The goal of the experiment was to develop a prototype 3D video game by a small development studio using created solutions. The experiment showed great possibilities of the developed tool and the technology used.
Platooning is one of the prospective applications, where autonomous driving and Vehicle-to-Vehicle and Vehicle-to-Infrastructure communications play a vital role. Various test proved that reduction of inter-car distance withing platoon of trucks results in significant fuel savings. Keeping of such short distance as well as awareness of the platooning cars about the vicinity assumes the presence of highly reliable sensors. In this paper, we briefly discuss the impact of sensor inaccuracy on the overall behaviour of the train-of-trucks driving in the autonomous fashion, supported by the presence of dedicated wireless system utilizing context information stored in databases and maps.
Imbalanced data classification is still remaining thje important topic and during the past decades, plenty of works are devoted to this field of study. More and more real-life based imbalanced class problems inspired researchers to come up with new solutions with better performance. Various techniques are employed such as data handling approaches, algorithm-level approaches, active learning approaches, and kernel-based methods to enumerate only a few. This work aims at applying a novel dynamic selection methods on imbalanced data classification problems. The experiments carried out on several benchmark datasets confirm its pretty high performance.
Applications used for spatial data processing can operate on files stored on servers, which is a commonly used approach. It is known that this method may not be the best fit for processing very large raster files, however, it is difficult to find useful information on the issue. This is the reason the authors of this article carried out research to determine the real impact of ICT infrastructure on processing of such files. The article also aims to verify whether utilizing virtual desktops can help mitigate the identified adverse effects on the process.
In this paper we propose solution of the problem of clinical gait analysis in post-stroke patients using advanced artificial intelligence approaches: fuzzy logic, neural networks, and fractal dimension. We focus on the stroke influence on gait pattern and features due to stroke is regarded one of the major causes of disability, including gait disorders. No doubt gait may be described by many parameters but it still needs advanced computational approach. Statistical analysis and simulation of gait features allow for relatively early detection of many limitations, selection of the proper therapeutic method, and assessment of the therapy progress. Results presented here are promising despite our approach needs for further studies toward clinical application.
As the arms race between the new kinds of attacks and new ways to detect and prevent those attacks continues, better and better algorithms have to be developed to stop the malicious agents dead in their tracks. In this paper, we evaluate the use of one of the youngest additions to the deep learning architectures, the Gated Recurrent Unit for its feasibility in the intrusion detection domain. The network and its performance is evaluated with the use of a well-established benchmark dataset, called NSL-KDD. The experiments, with the accuracy surpassing the average of 98%, proves that GRU is a viable architecture for intrusion detection, achieving results comparable to other state-of-the-art methods.
The article presents the key routing protocols for wireless mesh networks and routing metrics that affect transmission efficiency. Routing protocols are crucial element for performance of mesh networks. However, the specificity of the radio channel and transmission between closely neighboring nodes requires the use of routing techniques not yet implemented in wired networks. As part of the research includes a comparative analysis of representative routing protocols carried out in the OMNeT++ environment. Results of simulation has been presented and compared.
In this paper we study the downlink of an Orthogonal Frequency Division Multiplexing (OFDM) based cell that accommodates calls from different service-classes with different resource requirements. We assume that calls arrive in the system according to a Poisson process while the call admission is based on the Bandwidth Reservation (BR) policy. The BR policy is used for the reservation of subcarriers in favor of calls of high subcarrier requirements. To determine the most important performance metrics, i.e., Call Blocking Probabilities (CBP) and resource utilization in this system, we model it as a multirate loss model and propose recursive formulas which reduce the complexity of the calculations. The accuracy of the formulas is verified via simulation and found to be quite satisfactory.
In this paper examines effectiveness of HAAR feature-based cascade classifier for face detection in the presence of various image distortions. In the article we have focused on picture distortions that are likely to be met in everyday life, namely blurring, salt and pepper noise, contrast and brightness shifts and “fisheye” type distortion typical for wide-angle lens. In the paper present the mathematical model of the classifier and distortions, the training procedure and finally results of segmentation under various level of distortion. The test dataset is a large publicly available “Labelled Faces in the Wild” (LFW). Results show that Cascade Classifier finds it most difficult to recognize images that contain 70% noise type salt and pepper. The least impact on the effectiveness of the method use of blurred images even though the high parameter of blurring. From the obtained results it appears that the effectiveness of face detection is also affected by the adequate parameters of contrast and brightness.
Automatic classification of electronic elements based on image analysis may be useful for verification of a proper selection of electronic integrated circuits previously assembled in the Printed Circuit Boards (PCBs), as well as for an automated sorting of such elements in robotic systems supporting their production. Although modern high speed pick and place machines dedicated for small surface-mount devices (SMD), utilising very small packages, largely replaced the through-hole assembling technology in many industrial application, there are still some types of applications where such technology is less suitable, e.g. due to thermal, mechanical or power constraints. Hence, a proper classification of electronic elements in dual in-line packages (DIP) using shape analysis may be an important element for the combination with further steps based on the recognition of alphanumerical markings. Some experimental results obtained for selected shape descriptions are presented in this paper, which are promising also for natural images.
