This work proposes a supervised machine learning algorithm for target localization in deep brain stimulation (DBS). DBS is a recognized treatment for movement disorders, such as essential tremor. The effects of DBS significantly depends on the precise choice of target location. Recent research on diffusion tensor imaging (DTI) shows that the optimal target is related to the dentato-rubro-thalamic tract (DRTT), thus DRTT analysis has become a promising approach. This technique is more accurate than conventional ones, but still too complicated for clinical scenarios, where only magnetic resonance imaging (MRI) data is available. In order to improve efficiency and utility, we consider target localization as a non-linear regression problem in a reduced-reference learning framework, and solve it with convolutional neural networks. The proposed method is light-weight, and consists of two image-based networks: one for classification and the other for localization. We model the basic workflow as an image retrieval process and define relevant performance metrics. Using DRTT analysis as groundtruths, we show that DTI-based optimal targets can be inferred from MRI data with high accuracy. For 280x220 (0.7 mm slice thickness) MRI input, our model achieves an average posterior localization error of 2.3 mm, and a median of 1.7 mm. The proposed framework is the first in the DBS domain. It is a successful application of reduced-reference learning, and may serve as a baseline for general target localization problems in DBS.

Driving energy consumption plays an important role in the navigation of autonomous mobile robots in off-road scenarios. However, the accuracy of the driving energy predictions is often affected by a high degree of uncertainty due to unknown and constantly varying terrain properties, and the complex wheel-terrain interaction in unstructured terrains. In this paper, we propose a probabilistic deep meta-learning approach to model the existing uncertainty in the driving energy consumption and efficiently adapt the probabilistic predictions based on a small number of local measurements. Our method expands upon an existing deterministic deep-meta learning model that, in contrast, only provided single-point energy estimates. The performance of our method is compared against the deterministic approach in a 3D-body dynamic simulator over several typologies of deformable terrains and unstructured geometries. In this way, we provide evidence of the benefit of the proposed method to enhance the predictions with informative probabilistic considerations, which can be crucial to the safety of mobile robots traversing challenging, unstructured environments.

The solenoid has been used extensively in a wide range of electrical applications. In the present work, an analytical model is presented for estimating the 3-dimensional (3D) magneto quasi-static fields of the cylindrical solenoid with helix-shaped (helical) wire winding. The proposed integral formulation is based on the Schur (Hadamard) vector product of the current vector I and the helical tangent vector T, to express the magnetic vector potential A. The model is used to estimate the magnetic flux density vector B and the flux linkage Psi anywhere in 3D space, excluding the (source) regions of conducting wire. A constant scaling coefficient, based on measured terminal inductance, is proposed for calibration of the model’s magnetic field. Error analysis is presented on the numerical integration, based on truncated series expansion of complete elliptic integrals in terms of Chebychev polynomials. The model may be used to estimate all three cylindrical components of the magneto quasi-static fields of solenoids as a function of core diameter, wire radius, the number of wire turns per winding layer, the number of winding layers, and the complex permeability of frequency-dependent linear material. Several numerical examples are provided to validate the helical model against the superposition of circular loops, and an idealized circuit model.

Despite the advancement in deep learning-based semantic segmentation methods, which have achieved accuracy levels of field experts in many computer vision applications, the same general approaches may frequently fail in 3D medical image segmentation due to complex tissue structures, noisy acquisition, disease-related pathologies, as well as the lack of sufficiently large datasets with associated annotations. For expeditious diagnosis and quantitative image analysis in large-scale clinical trials, there is a compelling need to predict segmentation quality without ground truth. In this paper, we propose a deep learning framework to locate erroneous regions on the boundary surfaces of segmented objects for quality control and assessment of segmentation. A Convolutional Neural Network (CNN) is explored to learn the boundary related image features of multi-objects that can be used to identify location-specific inaccurate segmentation. The predicted error locations can facilitate efficient user interaction for interactive image segmentation (IIS). We evaluated the proposed method on two data sets: Osteoarthritis Initiative (OAI) 3D knee MRI and 3D calf muscle MRI. The average sensitivity scores of $0.96\pm0.00$ and $0.96\pm0.02$, and the average positive predictive values of $0.87\pm0.01$ and $0.93\pm0.03$ were achieved, respectively, for erroneous surface region detection of knee cartilage segmentation and calf muscle segmentation. Our experiment demonstrated promising performance of the proposed method for segmentation quality assessment by automated detection of erroneous surface regions in medical images.

Our aim is to develop next generation of MEMS (microelectromechanical systems) devices that are composed Ni-Mn-Ga and silicon layers. Due to the large magnetic-field-induced strains of Ni-Mn-Ga, actuating components can be fabricated in the Ni-Mn-Ga layers. Other functional components can be manufactured in the silicon layer. Single crystalline Ni-Mn-Ga alloys that are grown by using the Bridgman vertical growth technique have thus far obtained the largest magnetic field-induced strain (MFIS), a magnetic shape memory (MSM) effect. Similar to silicon wafers, Ni-Mn-Ga wafers are also sliced from crystal-oriented single crystalline ingots. To fabricate hybrid MEMS devices such as micromanipulators and robots, lab-on-chip containing micropump manifolds and valves, or vibration energy harvesters, the fabrication processes used for MEMS devices will be used are also be used to fabricate components in the Ni-Mn-Ga layer of the hybrid MEMS devices. One of the most important processes for MEMS fabrication is the structuring of materials by chemical etching. The main goal of this study is to obtain evidence that the etchant etches silicon but not Ni-Mn-Ga and to identify an etchant that etches Ni-Mn-Ga but not silicon. The present paper reports on a novel experiment in dissolving Ni-Mn-Ga alloys. An etchant composition of 69% HNO3, 98% H2SO4, and CuSO4•5H2O is proposed for dissolving Ni-Mn?Ga alloys and the variation in the dissolution rate by adjusting the concentrations of HNO3 and ultrapure water (UPW) is demonstrated. This etchant was demonstrated to etch Ni-Mn-Ga but not silicon. The HF+HNO3 acidic solution commonly used for etching silicon does not dissolve Ni-Mn-Ga

In recent years, electronic glasses, including augmented reality (AR), virtual reality (VR), and mixed reality (MR) devices that connect the natural world and virtual world seamlessly, have significantly developed. Ocular images are inherently acquired during the immersion experiences from such devices, and can enable the verification of privileged identities during a live broadcast or meetings in virtual spaces. Lack of any such public database, and any specialized framework, is one of the key challenges in advancing iris recognition capability in metaverse or such virtual spaces. We introduce first or a new public iris images database, from 384 different subjects, to advance iris recognition using a generalized AR/VR device. Conventional iris recognition methods can only offer limited performance on such challenging iris images. This paper introduces an accurate and generalizable framework for iris recognition using AR/VR devices. The proposed framework is based on a convolutional network that uses a specifically designed shifted and extended quadlet loss function, enabling the network to accurately learn the discriminant iris features preserved in close-range and off-angle iris images. The framework introduced in this work can also adaptively consolidate the spatially corresponding features and abstract features from the other ocular details for more accurate matching. Thorough experimental results presented in this paper, using several classical and state-of-art iris recognition methods, are consistently outperforming and validate the effectiveness of the proposed approach with improvement of 96.30%, 30.58% and 27.23% for true accept rate (at false accept rate =0.0001), and 85.65%, 49.91% and 76.56% for equal error rate respectively.

Federated learning provides data privacy protection by keeping data used for clients’ machine learning training private, and only sending model parameters updates to the centralised server/aggregator. However, the federated learning framework is still vulnerable to various attacks, such as data poisoning, launched by malicious/compromised clients. Cautious clients participating in federated learning, on the other hand, employ privacy protection techniques such as differential privacy to keep their model updates safe from inference attacks launched by the centralised aggregator. An aggregator thus needs to employ techniques to differentiate between model updates from benign, malicious and cautious clients, and to mitigate the effects of updates from clients other than benign clients. To reach this goal, we propose a novel federated learning system called FLAP which is robust against attacks launched by malicious clients and privacy protections employed by cautious clients.

A document by Martin Esugo . Click on the document to view its contents.

This article reports on the development and evaluation of a proposed optimal network reconfiguration approach for total real power losses reduction in radial distribution systems. The proposed approach was developed in MATLAB and the IEEE 33-bus radial distribution system was used in evaluating its performance. Application of the developed approach on the IEEE 33-bus radial distribution system helped to reduce the total real power losses of the system by 76.31%, from 202.68 kW to 48.02 kW, whereas the best approaches reported in literature reduced the losses by 31.15%, from 202.68 kW to 139.55 kW. These results suggest that the approaches reported in literature fail to find the global optimum combination of tie and sectionalizing branches whose corresponding switches have to be closed and opened, respectively, in order to minimize the total real power losses in the system. Consequently, due to its significant reduction of the system’s total real power losses, the developed approach proves to be a potentially reliable tool for power system operators to adopt and use when solving the radial tribution systems’ optimal network reconfiguration problem for real power losses reduction.

Millimeter-wave (mm-wave) frequency modulated continuous wave (FMCW) radars are increasingly being deployed for scenario perception in various applications. It is expected that the mutual interference between such radars will soon become a significant problem. Therefore, to maintain the reliability of the radar measurements, there must be procedures in place to mitigate this interference. This paper proposes a novel interference mitigation technique that utilizes the pulse compression principle for interference compression and mitigation. The interference in the received time-domain signal is compressed using an estimated matched filter. Afterward, the compressed interference is discarded, and the signal is repaired in the pulse-compressed domain using an autoregressive (AR) model. Since the interference spans fewer samples after compression, the signal can be restored more accurately in the compressed domain. Real outdoor measurements show that the interference is effectively suppressed down to the noise floor using the proposed scheme.

A document by Weixuan XIAO . Click on the document to view its contents.

Simulating such events typically involves a system of differential equations representing the overall generation and load present at the time. The standard model based on the swing equation assumes unlimited capacity in aggregated resources, as well as the availability of these services for the duration of the frequency excursion. In simulating the effect of outages on the GB Grid frequency on 2019/8/9, the effect of limiting these services to the capacity of resources engaged during the event is examined. It is shown that by taking these refinements into account the timing and extent of the frequency nadir can be successfully estimated. Insight is gained into the responses of various characteristics of the grid and how they interact with unplanned generation imbalances. Using this adapted model, further events on the GB grid are examined to validate the influence of these features.

Satellite-aerial-ground integrated network (SAGIN) has been widely envisioned as a promising network architecture for 6G. In the SAGINs, high altitude platform (HAP)-aided relaying satellite systems using hybrid free-space optics (FSO)/radio-frequency (RF) communications have recently attracted research efforts worldwide. Nevertheless, the main drawback of hybrid FSO/RF systems is the restricted bandwidth of the RF connection, especially when the FSO one is blocked by cloud coverage. This paper explores a novel solution for the hybrid FSO/RF HAP-based SAGIN under the impact of weather and atmospheric conditions. Specifically, an additional unmanned aerial vehicle (UAV) is deployed to diverse the FSO link from the HAP-to-ground station to avoid cloud blockage while maintaining a high-speed connection of the FSO link. A mirror array constructed by re-configurable intelligent surface (RIS), an emerging technology, is mounted on the UAV to reflect the signals from the HAP. The channel model of RIS-UAV takes into account both atmospheric turbulence and hovering-induced pointing errors. Furthermore, we present a novel link switching design with a multi-rate adaptation scheme for the proposed network under different weather and turbulence conditions. Numerical results quantitatively confirm the effectiveness of our proposal. Additionally, we provide insightful discussions that can be helpful for the practical system design of RIS-UAV-assisted HAP-based SAGIN using hybrid FSO/RF links. Monte Carlo simulations are also performed to validate the accuracy of theoretical derivations.

From the early stage of robotic applications to current autonomous driving technologies, environment modeling has been acting as the middleware of connecting perception and decision layers. In robotic applications, space-oriented models (e.g., grid map, drivable area) are widely applied to faithfully reflect the space occupation. With the development of autonomous driving, highly dynamic and complex road environment brings rising need of understanding the type and motion status of objects, thus element list became the mainstream of environment model. However, along comes the reliablity problem caused by missed detection and irregular objects, which is still inevitable despite the detection accuracy improvement. In view of this, we provide a new view of driving environment with the proposed unified state-extended boundary (USEB), aiming to improve the reliablity of element-oriented model. For driving decision requirements, different types of elements are consistently converted into driving constraints. Semantics and dynamics are expressed as status of drivable area boundary, making it possible to merge space occupation to improve reliability against missed detection and irregular objects. Evaluation of USEB is carried out on nuScenes dataset. Comparative results show that the proposed USEB could cover the required information of driving decision, whereas achieving higher reliability than commonly applied element-oriented model.

Wheezes are adventitious respiratory sounds commonly present in patients with respiratory conditions. The presence of wheezes and their time location are relevant for clinical reasons, such as understanding the degree of bronchial obstruction. Conventional auscultation is usually employed to analyze wheezes, but remote monitoring has become a pressing need during recent years. Automatic respiratory sound analysis is required to reliably perform remote auscultation. In this work we propose a method for wheeze segmentation. Our method starts by decomposing a given audio excerpt into intrinsic mode frequencies using empirical mode decomposition. Then, we apply harmonic-percussive source separation to the resulting audio tracks and get harmonic-enhanced spectrograms, which are processed to obtain harmonic masks. Subsequently, a series of empirically derived rules are applied to find wheeze candidates. Finally, the candidates stemming from the different audio tracks are merged and median filtered. In the evaluation stage, we compare our method to three baselines on the ICBHI 2017 Respiratory Sound Database, a challenging dataset containing various noise sources and background sounds. Using the full dataset, our method outperforms the baselines, achieving an F1 of 41.9%. Our method’s performance is also better than the baselines across several stratified results focusing on five variables: recording equipment, age, sex, body-mass index, and diagnosis. We conclude that, contrary to what has been reported in the literature, wheeze segmentation has not been solved for real life scenario applications. Adaptation of existing systems to demographic characteristics might be a promising step in the direction of algorithm personalization, which would make automatic wheeze segmentation methods clinically viable.

We present a rigorous semi-analytical formulation for the analysis of electromagnetic (EM) scattering from a cylinder constructed from a metasurface represented using surface susceptibilities. The formulation uses the Generalized Sheet Transition Conditions (GSTCs) to represent the surface and eigenfunction expansion (EFE) of the incident and scattered fields, by exploiting their angular periodicity. Incorporating a completely general non-uniform surface formulation of 36 susceptibility components, a matrix equation is formulated that can be solved for the field harmonic coefficients. The paper illustrates the methodology with a number of examples; including two formed from a finite sized practical unit-cell exhibiting a normally oriented magnetic resonance which is compared with commercial EM full-wave solver; other examples present surfaces that have a modulated gain/loss profile (i.e. amplitude modulation) and polarization conversion. It is found that for all cases the EFE solution very accurately captures the scattered fields in both the interior and exterior regions of the surface. Detailed convergence studies are further presented including the effect of susceptibility modulation on the number of terms needed in the EFEs, to reach a correct field solution. The proposed EFE framework is further compared with a second GSTC based simulator using an Integral Equations (IE) approach and it is found that the IE-GSTC simulated fields approach that of the EFE as the surface discretization is increased. The proposed EFE approach is thus a quick, rigorous methodology which, although limited in the geometry which it can model, has many advantages for investigation into the use of GSTCs, application development and providing a baseline for simulation studies.

Autonomous vehicle requires a general planner for all possible scenarios. Existing researches design such a planner by a unified scenario description. However, it may significantly increase the planner complexity even in some simple tasks, e.g., car following, further resulting in unsatisfactory driving performance. This work aims to design a general planner which can 1) drive in all possible scenarios and 2) have lower complexity in some common scenarios. To this end, this work proposes a pertinent boundary for multi-scenario driving planning. The total approach is named as Pertinent Boundary-based Unified Decision system. Based on the original drivable area, the pertinent boundary can further support motion status and semantics of the traffic elements, which provides the potential of pertinent performance for given scenarios. The pertinent boundary can support unified driving with the drivable area, in the meantime, can be pertinently modified to support the pertinent driving decisions for identified driving scenarios (e.g., car-following, junction left turning). It will further avoid the bump between the connections of the scenarios due to the continuity of space boundary. Thus, the planner is suitable for the fully autonomous driving. The proposed method is validated in different classical driving decision scenarios. Results show that the proposed method can support pertinent driving decisions in identified scenarios, in the meantime, assure generalized cross-scenario planning when no scenario information is available. Such a method shed light on fully autonomous driving by pertinence improvement of multi-scenario decision in the complex real world.

Deep learning techniques are very prominent in processing remotely sensed synthetic aperture radar (SAR) images for real-time, high-impact applications, such as image classification, object detection, and semantic segmentation. The accuracy of deep learning models, such as convolutional neural networks (CNNs), depends on the quality of the input data. Compared to the model-centric approach, where the model parameters are optimized during training, the data-centric approach can enhance the performance accuracy as data quality is improved before training the models. Improving the data quality of SAR images is challenging as SAR image properties are different from optical (OPT) images. Image fusion techniques proved to enhance the quality of SAR images when combined with OPT images. Many fusion techniques exist for combining SAR and OPT images in the classical domain. This paper proposes a novel approach to using quantum computing for the image fusion of SAR and OPT images. Eight different quantum processing techniques are used for the fusion of the images. We designed and created a dataset for land-use classification by collecting data using the Google Earth Engine. The quality metric measurements show that the quality of SAR images has improved by using the proposed quantum processing techniques. In addition, performance evaluation of the deep learning CNNs on the dataset was carried out for all quantum processing techniques. Our approach improved the classification accuracy from 82.64%, with only SAR images for training, to 95.36% using the proposed image fusion techniques.