ETRI Journal最新文献

Different diseases occur in the brain. For instance, hereditary and progressive diseases affect and degenerate the white matter. Although addressing, diagnosing, and treating complex abnormalities in the brain is challenging, different strategies have been presented with significant advances in medical research. With state-of-art developments in artificial intelligence, new techniques are being applied to brain magnetic resonance images. Deep learning has been recently used for the segmentation and classification of brain images. In this study, we classified normal and pathological brain images using pretrained deep models through transfer learning. The EfficientNet-B5 model reached the highest accuracy of 98.39% on real data, 91.96% on augmented data, and 100% on pathological data. To verify the reliability of the model, fivefold cross-validation and a two-tier cross-test were applied. The results suggest that the proposed method performs reasonably on the classification of brain magnetic resonance images.

We propose an adaptive unmanned aerial vehicle (UAV)-assisted object recognition algorithm for urban surveillance scenarios. For UAV-assisted surveillance, UAVs are equipped with learning-based object recognition models and can collect surveillance image data. However, owing to the limitations of UAVs regarding power and computational resources, adaptive control must be performed accordingly. Therefore, we introduce a self-adaptive control strategy to maximize the time-averaged recognition performance subject to stability through a formulation based on Lyapunov optimization. Results from performance evaluations on real-world data demonstrate that the proposed algorithm achieves the desired performance improvements.

An ultrawide stopband lowpass filter is reported using three stepped impedance resonators with high selectivity. The filter extends the stopband frequency range and attenuation, and two quarter-wave open stubs and three circular ground slots are introduced. The lumped and distributed equivalent models are derived and analyzed. The corresponding results are validated experimentally in a fabricated prototype. The prototype lowpass filter has a 3 dB cutoff frequency (f_c) of 2.9 GHz, and the stopband is extended up to 35 GHz (12.07f_c), with an attenuation level better than 20 dB throughout. The passband-to-stopband transition (3 dB–20 dB) bandwidth is 0.18 GHz, and the roll-off factor is 135 dB/GHz at 30 dB. The insertion loss is 0.3 dB at 1.6 GHz. The normalized circuit size of the proposed filter with respect to the guided wavelength is 0.04.

In vehicular networks, diverse safety information can be shared among vehicles through internet connections. In vehicle-to-internet communications, vehicles on the road are wirelessly connected to different cloud networks, thereby accelerating safety information exchange. Onboard sensors acquire traffic-related information, and reliable intermediate nodes and network services, such as navigational facilities, allow to transmit safety information to distant target vehicles and stations. Using vehicle-to-network communications, we minimize delays and achieve high accuracy through consistent connectivity links. Our proposed approach uses intermediate nodes with two-hop separation to forward information. Target vehicle detection and routing of safety information are performed using machine learning algorithms. Compared with existing vehicle-to-internet solutions, our approach provides substantial improvements by reducing latency, packet drop, and overhead.

This study proposes a novel, conditionally applied neural network technique to reduce the overall peak-to-average power ratio (PAPR) of an orthogonal frequency division multiplexing (OFDM) system while maintaining an acceptable bit error rate (BER) level. The main purpose of the proposed scheme is to adjust only those subcarriers whose peaks exceed a given threshold. In this respect, the developed C-ANN algorithm suppresses only the peaks of the targeted subcarriers by slightly shifting the locations of their corresponding frequency samples without affecting their phase orientations. In turn, this achieves a reasonable system performance by sustaining a tolerable BER. For practical reasons and to cover a wide range of application scenarios, the threshold for the subcarrier peaks was chosen to be proportional to the saturation level of the nonlinear power amplifier used to pass the generated OFDM blocks. Consequently, the optimal values of the factor controlling the peak threshold were obtained that satisfy both reasonable PAPR reduction and acceptable BER levels. Furthermore, the proposed system does not require a recovery process at the receiver, thus making the computational process less complex. The simulation results show that the proposed system model performed satisfactorily, attaining both low PAPR and BER for specific application settings using comparatively fewer computations.

Telecommunications through an electrically conductive medium require the use of carrier bands with very-low and ultralow frequencies to establish radiofrequency links in harsh environments. Recent advances in atomic magnetometers operating at very-low frequencies have facilitated the reception of digitally modulated signals. We demonstrate the transmission and reception of quadrature phase-shift keying (QPSK) signals using a multi-resonant loop antenna and atomic magnetometer, respectively. We report the measured error vector magnitude according to the symbol rate for QPSK modulation and analyze the bandwidth of a receiver based on the atomic magnetometer. The multi-resonant loop antenna noticeably enhances the bandwidth by over 70% compared with a single-loop antenna. QPSK modulation for a carrier frequency of 20 kHz and symbol rate of 150 symbols per second verifies the feasibility of demodulation, and the measured error vector magnitude and signal-to-noise ratio are 7.29% and 30.9 dB, respectively.

This paper surveys recent multiagent reinforcement learning and neural Myerson auction deep learning efforts to improve mobility control and resource management in autonomous ground and aerial vehicles. The multiagent reinforcement learning communication network (CommNet) was introduced to enable multiple agents to perform actions in a distributed manner to achieve shared goals by training all agents' states and actions in a single neural network. Additionally, the Myerson auction method guarantees trustworthiness among multiple agents to optimize rewards in highly dynamic systems. Our findings suggest that the integration of MARL CommNet and Myerson techniques is very much needed for improved efficiency and trustworthiness.

Recent semantic segmentation frameworks usually combine low-level and high-level context information to achieve improved performance. In addition, postlevel context information is also considered. In this study, we present a Context ReFinement Network (CRFNet) and its training method to improve the semantic predictions of segmentation models of the encoder–decoder structure. Our study is based on postprocessing, which directly considers the relationship between spatially neighboring pixels of a label map, such as Markov and conditional random fields. CRFNet comprises two modules: a refiner and a combiner that, respectively, refine the context information from the output features of the conventional semantic segmentation network model and combine the refined features with the intermediate features from the decoding process of the segmentation model to produce the final output. To train CRFNet to refine the semantic predictions more accurately, we proposed a sequential training scheme. Using various backbone networks (ENet, ERFNet, and HyperSeg), we extensively evaluated our model on three large-scale, real-world datasets to demonstrate the effectiveness of our approach.

Various techniques for noninvasively focus microwave energy on lesions have been proposed for thermotherapy. To focus the microwave energy on the lesion, a focusing parameter, which is referred to as the magnitude and phase of microwaves radiated from an external array antenna, is very important. Although the finite-difference time-domain (FDTD)-based time-reversal (TR) focusing algorithm is widely used, it has a long processing time if the focusing target position changes or if optimization is needed. We propose a technique to obtain multistatic data (MSD) based on Green's function and use it to derive the focusing parameters. Computer simulations were used to evaluate the electric fields inside the object using the FDTD method and Green's function as well as to compare the focusing parameters using FDTD- and MSD-based TR focusing algorithms. Regardless of the use of Green's function, the processing time of MSD-based TR focusing algorithms reduces to approximately 1/2 or 1/590 compared with the FDTD-based algorithm. In addition, we optimize the focusing parameters to eliminate hotspots, which are unnecessary focusing positions, by adding phase-reversed electric fields and confirm hotspot suppression through simulations.