Search

Data collected by Earth-observing (EO) satellites are often afflicted by cloud cover. Detecting the presence of clouds -- which is increasingly done using deep learning -- is crucial preprocessing in EO applications. In fact, advanced EO satellites perform deep learning-based cloud detection on board the satellites and downlink only clear-sky data to save precious bandwidth. In this paper, we highlight the vulnerability of deep learning-based cloud detection towards adversarial attacks. By optimising an adversarial pattern and superimposing it into a cloudless scene, we bias the neural network into detecting clouds in the scene. Since the input spectra of cloud detectors include the non-visible bands, we generated our attacks in the multispectral domain. This opens up the potential of multi-objective attacks, specifically, adversarial biasing in the cloud-sensitive bands and visual camouflage in the visible bands. We also investigated mitigation strategies against the adversarial attacks. We hope our work further builds awareness of the potential of adversarial attacks in the EO community.

Quantum annealing is a promising paradigm for building practical quantum computers. Compared to other approaches, quantum annealing technology has been scaled up to a larger number of qubits. On the other hand, deep learning has been profoundly successful in pushing the boundaries of AI. It is thus natural to investigate potentially game changing technologies such as quantum annealers to augment the capabilities of deep learning. In this work, we explore binary neural networks, which are lightweight yet powerful models typically intended for resource constrained devices. Departing from current training regimes for binary networks that smooth/approximate the activation functions to make the network differentiable, we devise a quadratic unconstrained binary optimization formulation for the training problem. While the problem is intractable, i.e., the cost to estimate the binary weights scales exponentially with network size, we show how the problem can be optimized directly on a quantum annealer, thereby opening up to the potential gains of quantum computing. We experimentally validated our formulation via simulation and testing on an actual quantum annealer (D-Wave Advantage), the latter to the extent allowable by the capacity of current technology., Comment: 13 pages, 4 figures

Generalised zero-shot learning (GZSL) methods aim to classify previously seen and unseen visual classes by leveraging the semantic information of those classes. In the context of GZSL, semantic information is non-visual data such as a text description of the seen and unseen classes. Previous GZSL methods have explored transformations between visual and semantic spaces, as well as the learning of a latent joint visual and semantic space. In these methods, even though learning has explored a combination of spaces (i.e., visual, semantic or joint latent space), inference tended to focus on using just one of the spaces. By hypothesising that inference must explore all three spaces, we propose a new GZSL method based on a multimodal classification over visual, semantic and joint latent spaces. Another issue affecting current GZSL methods is the intrinsic bias toward the classification of seen classes - a problem that is usually mitigated by a domain classifier which modulates seen and unseen classification. Our proposed approach replaces the modulated classification by a computationally simpler multidomain classification based on averaging the multi-modal calibrated classifiers from the seen and unseen domains. Experiments on GZSL benchmarks show that our proposed GZSL approach achieves competitive results compared with the state-of-the-art.

Voltage transients are the main causes of complaints from medium-voltage customers. For this reason, utilities and independent authorities are focusing their attention on investigation techniques capable of providing information useful for classifying voltage transients depending on their origin. Distributed measurement systems for power quality monitoring able to detect, measure, and classify impulsive transients can be used in coordination with lightning location systems to assess whether there is a relation between a given transient and a lightning event. In this paper, a criterion for evaluating the above correlation is proposed and a numerical example, which is based on real accuracy specifications, is presented. The results show that attempting a correlation between lightning events and induced transients is quite reasonable. © 2004 IEEE.

The problem of determining the location of a fault in an electrical power network can be faced by using a distributed measurement system. This paper deals with the metrological characterization of the remote units of the system, so that the combined uncertainty affecting the fault position can be evaluated. A numerical approach has been used in this paper for the uncertainty estimation. Results are presented and discussed.

The use of a distributed measurement system to locate the source of transient voltages is discussed in this paper. The system implements a method based on the measurement of the starting instants of a transient in the nodes of the monitored network. The method performance is investigated by simulating the system operation. Results in the case of both faults and insertion of capacitors in a IEEE test network are presented and discussed.

There is a search for investigation techniques capable of characterizing voltage transients, which are one of the main causes of power quality degradation, and classifying them depending on their origin. In a previous paper, the use of a distributed measurement system for power quality monitoring in parallel with a lightning location system was proposed to assess whether a lightning event is responsible for a voltage transient. An accurate estimate of the transient's starting instant is required to tackle that task. In this paper, a system for the measurement of the starting instant of impulsive voltage transients is proposed and the procedures for its metrological characterization are described. The results of the characterization tests are also reported and discussed.

The offer provided by an electric utility combines many characteristics of a service together with the continuous delivery of electricity as a product. The quality of electricity as a processed material, resulting from a continuous process and having immediate delivery in continuous flow can he adversely affected by the end-use processes of other customers, as well as by the continuous-process reliability. Hence, the guidelines endorsed by ISO Standard 9004-3 for the quality management of processed materials must be suitably tailored in the case of electricity. Problems that must be dealt with to control the characteristics critical to quality of electricity are discussed. In this connection, measurement activities oriented to quality management are considered, mainly to outline unsolved problems.

Owing to the proliferation of loads which draw nonsinusoidal currents, the current and the voltage distortion in public distribution systems is ever increasing. The usual online investigation methods cannot provide information about the responsibility for harmonic distortion in a given metering section. New methodologies are therefore searched for. Some recently proposed investigation techniques are recalled in this paper and the relevant measurement problems are discussed. A digital measurement system which is based on the virtual instrumentation (VI) technique and implements these methods, is described. The results of some experimental work are also presented.

Modern instrumentation for the monitoring of electrical power systems is based on digital signal processing (DSP) techniques. If suitable design choices are made, the overall accuracy of this kind of instrumentation basically depends on the signal conditioning circuits. They have outputs proportional to the input to drive the A/D converters and are very often based on instrument transformers to ensure the isolation between the instrumentation and the monitored power system. This paper presents a lab prototype of a digital instrument for the calibration of current-to-voltage transducers (CVT's) and discusses its performance. The results of calibration tests performed on some commercial active and passive CVT's are also reported.

The use of digital signal processing techniques allows measuring any quantity mathematically defined. However, the application of the rules endorsed by the ISO "Guide to the expression of uncertainty in measurement" to evaluate the measurement uncertainty can turn into a difficult task, for instance, when quantities different from those usually measured to characterize the operation of power systems must be determined. In this paper, a numerical approach is proposed to easily tackle this task. Contrary to the closed-form approach, which allows to evaluate the output uncertainty by analytically solving the uncertainty propagation law, as prescribed by the ISO Guide, the method proposed numerically finds the uncertainty by applying a recursive algorithm. First, first the validity of the numerical approach is proved in the case of the spectral analysis of a periodic signal, which is the first step for nonconventional measurement. Then, the approach is used to evaluate the uncertainty in the measurement of the harmonic losses of a power transformer supplying nonsinusoidal load currents. The results confirm that these losses can be accurately determined from the difference between two powers measured as correlated quantities at the transformer output terminals.