In this paper, we present an accurate design approach for a class-E² wireless power transfer (WPT) system to enhance efficiency. The design takes into account the effects of nonlinear parasitic capacitances in switching devices, linear parasitic capacitances of coupling coils, and the forward drop voltage of the diode on the system's performance. To address the discontinuities in the measured data of nonlinear parasitic capacitances in switching devices, spline interpolation is used. A design example is also given along with its simulation and experimental results. All the switch-voltage waveforms meet the class-E zero-voltage-switching/zero-derivative-switching (ZVS/ZDS) conditions. Moreover, the simulated and experimental results closely match the numerical predictions, confirming the validity of the proposed design given in this paper.

We present an analysis-based design method for the class-E² wireless power transfer (WPT) system, which accounts for the parasitic components of switching devices, such as the on-resistances of MOSFETs and diodes, as well as the forward voltage drops of diodes, across any duty ratio. A prototype is designed and tested to validate the proposed method. The simulated and experimental results demonstrate good agreement with analytical predictions. Moreover, all switch voltage waveforms satisfy the class-E zero-voltage-switching/zero-derivative-switching (ZVS/ZDS) conditions, highlighting the impact of parasitic components in switching devices and confirming the effectiveness of the proposed design method.

In this study, we design and simulate two coupled chaotic circuits with memristors. As a result, we obtain nonlinear phenomena of the phase differences from coupled chaotic circuits with fixed coupling strength. These are new phenomena in which the phase differences change periodically or chaotically with time.

Physically Unclonable Functions (PUFs) are security primitives designed to generate chip-specific responses by exploiting inherent device variations that are difficult to reproduce. One approach to implementing a PUF is known as Response Generation according to Delay Time Measurement (RG-DTM), which measures the delay time difference between two paths using a multi-offset sense amplifier. However, implementing the RG-DTM PUF on a Field-Programmable Gate Array (FPGA) is impractical, as the sense amplifier is an analog circuit. To address this limitation, the fDTM-PUF has been proposed, which uses multiple D flip-flops (DFFs) to measure the delay time differences. Despite this improvement, implementing the fDTM-PUF on an FPGA remains challenging due to the strict layout constraints of FPGAs. In this paper, we propose the fTDC-PUF, which measures the delay time difference using a Time-to-Digital Converter (TDC) sensor. The fTDC-PUF is more easily implemented on FPGAs and allows for finer delay time measurements compared to conventional methods. This paper evaluates the fundamental characteristics of the fTDC-PUF, including uniqueness and steadiness, under various settings.

The rapid advancement of generative AI has been driven by the introduction of the Transformer model [1] [2]. However, a significant challenge with such models lies in their high computational requirements. As a result, generative AI predominantly relies on online services. Nevertheless, exploring generative AI capabilities on the edge holds significant potential for enabling novel applications. By employing lightweight models, it becomes feasible to design architectures specifically for edge applications.

Federated Learning (FL) enables distributed learning by sharing model parameters while preserving data privacy. While it proves effective for single tasks, its application to multiple tasks remains challenging. This study proposes Adaptive Tuning of Layer Sharing (ATLAS), a method that dynamically selects shared layers and optimizes task weighting based on task similarity. ATLAS improves the efficiency and accuracy of multi-task FL while reducing communication costs.

In this paper, we propose an information communication method that combines a multimodal communication function that encompasses reception by the human eye and reception by the device eye, and a multidirectional communication function that encompasses multidirectional and multiviewpoint communication. This system uses a multimodal information communication system that uses devices and human vision as receivers and the refraction of lenses to provide information from multiple viewpoints. The communication area and the symbol error rate (SER) are evaluated.

In this paper, the average retention time, average recovery time, symbol error rate, and effective data transmission efficiency taking account of the synchronization performance and the symbol error rate in a synchronization method using two binary pulse position modulation (BPPM) signals of adjacent modified multipulse pulse position modulations (modified MPPMs) have been evaluated. It is found that there is an optimum value for the number of pulses generated in the modified MPPM frame in terms of synchronization performance, the symbol error rate, and the effective data transmission efficiency characteristics.

Rescue workers in disasters and operators in construction sites become fatigued because they perform demanding daily tasks. Such fatigue causes discomfort and a lack of motivation at work, and these factors can have a serious impact on work efficiency and safety. Therefore, an approach is required to constantly monitor and detect fatigue conditions accurately. A head-mounted camera refers to a camera attached to a helmet, and it is used to record video as needed in order to share the situation at the work site during daily tasks by rescue workers and operators. Additionally, a head-mounted camera allows for hands-free recording. In this paper, a human-recorder device is proposed, which can detect fatigue by recording video with a head-mounted camera. The human-recorder extracts head sway from recorded videos and performs frequency analysis on this sway to detect fatigue. In the experiment, subjects perform an n-back task for 20 minutes to cause fatigue. From this result, the frequency analysis of head sway revealed a peak frequency between 1 ∼ 2 Hz. This peak frequency enables the estimation of the subject's pulse rate. In addition, it can be observed that low frequency components (0 ∼ 1 Hz) increase or decrease after the task. Therefore, the human-recorder can detect the fatigue by increasing or decreasing low frequency.

Quantitative assessment of chewing ability is essential in healthcare. Current methods, such as using glucose-containing gummies, face challenges in measurement. To mitigate these challenges, a previous study introduced a technique to estimate the amount of glucose extracted based on gummy chewing sounds. In this study, we modify this existing method using a large-scale self-supervised learning model called wavLM to enhance accuracy. Validation on simulated data demonstrated remarkable accuracy improvements using the encoder output of wavLM combined with a one-dimensional convolutional neural network.

This study aims to develop an anomaly detection method for cancer in positron emission tomography (PET) images. Our database consists of 1,360 maximum-intensity projection (MIP) images generated from whole-body PET volumes, including 1,030 normal cases and 330 abnormal cases. A self-attention mechanism and normalizing flow are introduced into PatchCore for anomaly detection. The self-attention mechanism is used to consider the feature vectors across the entire image, whereas the normalizing flow transforms the distributions of the feature vectors in the feature maps to approximate the standard normal distribution. Each MIP image is classified as abnormal or normal based on an anomaly score, which is defined as the Euclidean distance to the nearest feature vector in the memory bank of PatchCore. With the proposed method, the classification accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve were 73%, 71%, 75%, and 0.77, respectively, showing improvements compared with the conventional PatchCore (67%, 63%, 70%, and 0.72, respectively). The proposed method could be useful for identifying cancerous lesions and reducing the interpretation time of PET screening.

This study aims to develop a computerized attenuation correction method for liver single photon emission computed tomography (SPECT) images using an extended 3D-pix2pix model with bidirectional learning. Our database consisted of liver SPECT images obtained from 809 patients. For each patient, SPECT images were reconstructed both with and without CT-based attenuation correction. In this study, the conventional 3D-pix2pix architecture is extended by incorporating a conditional encoder and two decoders, allowing for mutual transformation between SPECT images before and after attenuation correction. The conditional encoder takes an input image along with a class label indicating whether the input image is pre- or post-attenuation correction. This shared encoder facilitates the extraction of consistent features from both pre- and post-attenuation corrected SPECT images. The two decoders generate virtual post- or virtual pre-attenuation corrected image, respectively, from the input pre- or post-attenuation corrected image. With the proposed method, the root mean squared error (RMSE) and structural similarity index measure (SSIM) were 15.94 and 0.916, respectively, showing higher fidelity compared to the conventional 3D-pix2pix model (16.81 and 0.913). The proposed attenuation correction method for liver SPECT images achieved high fidelity and shows potential usefulness as an attenuation correction method without CT.

The Japan Aerospace Exploration Agency (JAXA) aims to develop an image navigation system to accurately estimate the self-position of space probes. In recent years, extensive research has been conducted on image recognition using machine learning. Object detection and homography matrix estimation methods based on convolutional neural networks (CNN) exhibit high discrimination and generalization performance. However, implementing these systems on small-space probes is challenging because of enormous computational resource requirements. This paper proposes a lunar crater recognition system using binary neural networks (BNN) with low-computation resource hardware. An overview of the system is provided along with validation results. The proposed method was demonstrated to be effective in lunar crater recognition tasks on low-computation resource hardware, allowing the implementation of a CNN-based lightweight system in such an environment.

Bumblebees are often used as pollinators in greenhouse cultivation. However, they become less active in hot environments, such as during summer. Additionally, the rising procurement costs owing to the mass mortality of bumblebees have become a significant issue. To address these challenges, we propose an artificial pollination system that uses small drones as an alternative to bees. In this study, we integrate the success-history-based adaptive differential evolution (SHADE) algorithm into the proposed system to optimize the drone's path control and reduce power consumption during flower searching. The algorithm is modified to enhance its suitability for the proposed system. It is specifically designed to minimize hovering time, a key factor in drone power consumption. Furthermore, a simulated greenhouse environment was used to determine whether the drones employed swarm intelligence or an evolutionary algorithm for flower searching. The enhanced SHADE algorithm, adapted for this method, enables efficient and accurate flower detection while significantly reducing energy consumption compared to conventional SHADE and differential evolution algorithms.

We propose a cloud image prediction method using convolutional LSTM to improve the prediction accuracy for photovoltaic power generation systems. In particular, we focus on the effects of input data of the prediction model to improve the accuracy of cloud movement predictions. Future cloud images are generated (estimated) using cloud movement vectors obtained from time-series cloud images. Then, we examine the future images as input data for the model. The performance of the proposed method is evaluated from the error and correlation between the ground truth image and the predicted image.

We propose a bin picking system for objects using a position and pose recognition method based on AI image processing. The image of objects is captured by a monocular camera and the position and pose of objects in the image are recognized using YOLOv8. Using the recognition results, the picking motion is determined, and the robot arm (DOBOT Magician) executes the bin picking task. To evaluate the object recognition accuracy and motion control stability of the bin picking system, the determined motions were executed multiple times on randomly arranged bolts in an experiment.

We propose an obstacle detection and road surface recognition system utilizing a depth camera installed in a position looking down on the electric wheelchair. The system detects grooves by comparing the depth information obtained from the depth camera with the ground level. However, the detection results may include noise or safe areas such as drainage holes [1]. The system employs morphological transformations to remove small regions, taking into account the contact area of the front wheels of the wheelchair. The electric wheelchair is then automatically stopped in accordance with the detection results to prevent accidents. The effectiveness of the proposed system was verified through driving experiments using an electric wheelchair.