Search

Objective: Based on software engineering principles, we designed and implemented a medical device data dictionary system that aligns with data integration and industry standards to fulfill the data management needs of medical institutions and healthcare professionals. Materials and Methods: Utilizing the web-based framework alongside open-source technologies such as Spring Boot, Spring Security, Apache MyBatis, Vue, and Element UI, along with database technologies such as MySql, MongoDB, and Redis, we analyze the basic elements of medical device data dictionary system and incorporate layered architecture technology, design the system architecture, functional workflows and database model required for the dictionary system, and conduct professional testing and trial evaluation of the system. Results and Analysis: In this study, we constructed a web-based open medical device data dictionary system that supports front-end and back-end operations. Using a ventilator as a sample for non-integrity assessment, the results of the system utility test showed that the system is functionally complete, satisfies the compatibility requirements of mainstream browsers and poses no significant risks or vulnerabilities to system operation. Conclusion: A medical device data dictionary that characterizes life cycle usage attributes based on their structural and functional properties was constructed. It provides a unified and standardized base for the management and analysis of medical device data, and helps realize the interaction and exchange of medical device data between disparate medical information systems.

Open source component libraries are widely used in software development. However, vulnerabilities of these components can threaten software security as they spread. Therefore, the industry commonly uses Software Composition Analysis (SCA) technology to detect the security of open source components in software. However, due to the low effectiveness of feature selection and the difficulty of precise feature extraction from open source component libraries, the accuracy of component analysis is not high. In this paper, we propose a feature extraction method for open source component libraries of binary programs based on fingerprint analysis. The fingerprint library is constructed based on 30,000 open source projects on the GitHub platform. It is proposed to use exported function fingerprint analysis, binary compilation fingerprint analysis, source code strings, etc. to extract the component library. With the fingerprint, we achieved precise positioning of the open source component library of binary programs, and developed the prototype tool Csrcc Sca, which achieved remarkable results by testing and evaluating 164 firmware packages related to intelligent connected vehicle. In terms of component version identification of component software packages, an accuracy rate of up to 96.81% was achieved; in terms of component layout identification of firmware packages, the accuracy rate also reached 83.33%.

Quality of GaN LEDs is closely related to the defects in quantum wells. Effectively evaluating the performance of LEDs is a key step in mass production. Electrical characterization testing is very effective for characterizing the performance of LEDs, and the I-V curve is an important evaluation parameter. In this paper, we analyzed correlation in photoluminescence (PL) spectra and I-V curves among different LEDs and studied the feasibility of using PL to predict and estimate the characteristics of the I-V curves. Firstly, in the experiment, we varied the excitation light power and tested the photoluminescence (PL) spectral characteristics of different samples. Subsequently, I-V tests were conducted on these samples with current injection. The results of the I-V curve tests indicated significant differences among different samples, which were consistent with the PL results. Corresponding carrier transport model was established, and the I-V curves were fitted, resulting in different ideality factors for those samples. The results indicated that samples with lower photoluminescence intensity exhibited a larger ideality factor in the I-V curves. On the contrary, for samples with higher photoluminescence intensity, the resulting ideality factor is smaller, approaching 2. A deeper analysis reveals a negative correlation between the relative strength of the normal peak and the abnormal peak in photoluminescence spectra and the value of the ideality factor. In the future, based on this correlation, photoluminescence-based detection techniques can be used to predict the I-V curve characteristics of LEDs in scenarios where direct current injection is not feasible.

Objective: Pulmonary cavity lesion is one of the commonly seen lesions in lung caused by a variety of malignant and non-malignant diseases. Diagnosis of a cavity lesion is commonly based on accurate recognition of the typical morphological characteristics. A deep learning-based model to automatically detect, segment, and quantify the region of cavity lesion on CT scans has potential in clinical diagnosis, monitoring, and treatment efficacy assessment. Methods: A weakly-supervised deep learning-based method named CSA2-ResNet was proposed to quantitatively characterize cavity lesions in this paper. The lung parenchyma was firstly segmented using a pretrained 2D segmentation model, and then the output with or without cavity lesions was fed into the developed deep neural network containing hybrid attention modules. Next, the visualized lesion was generated from the activation region of the classification network using gradient-weighted class activation mapping, and image processing was applied for post-processing to obtain the expected segmentation results of cavity lesions. Finally, the automatic characteristic measurement of cavity lesions (e.g., area and thickness) was developed and verified. Results: the proposed weakly-supervised segmentation method achieved an accuracy, precision, specificity, recall, and F1-score of 98.48%, 96.80%, 97.20%, 100%, and 98.36%, respectively. There is a significant improvement (P < 0.05) compared to other methods. Quantitative characterization of morphology also obtained good analysis effects. Conclusions: The proposed easily-trained and high-performance deep learning model provides a fast and effective way for the diagnosis and dynamic monitoring of pulmonary cavity lesions in clinic. Clinical and Translational Impact Statement: This model used artificial intelligence to achieve the detection and quantitative analysis of pulmonary cavity lesions in CT scans. The morphological features revealed in experiments can be utilized as potential indicators for diagnosis and dynamic monitoring of patients with cavity lesions

Hyperspectral image (HSI) super-resolution (SR) in both spatial and spectral dimensions is one of the most attractive research topics in HSI processing. Although recent advances in deep learning (DL) frameworks have greatly improved the performance of spatial-spectral SR reconstruction, existing methods learn discrete representations of HSI, ignoring real-world signals' continuous nature. Recently, Implicit Neural Representation (INR) has been applied to 3D surface reconstruction and image SR for continuous representation and has attracted increasing attention. In this paper, we propose the Local Implicit Spatial-spectral Function (LISSF), which learns a local continuous representation of high spatial resolution hyperspectral images (HR-HSI) from the discrete inputs. The model consists of a deep feature encoder and a spatial-spectral intensity decoder. The encoder converts the low spatial resolution multispectral image (LR-MSI) into deep features and the decoder predicts the intensity values at the given coordinates as output. Since the spatial-spectral coordinates are continuous, LISSF can achieve spatial-spectral SR in arbitrary scales, even extrapolating to higher resolutions not covered by the training data. Extensive experiments on spatial-spectral SR, spatial SR, and spectral SR demonstrate that LISSF can achieve superior performance in comparison with state-of-the-art methods. Moreover, ablation studies are performed on the effects of individual components of LISSF.

Irregular mare patch (IMP) is one of the most enigmatic volcanic features in the lunar nearside maria. Their apparently fresh surface morphology and derived young model ages contradict the currently proposed thermal evolution history of the Moon. Lunar orbital synthetic aperture radar (SAR) data of Mini-RF onboard Lunar reconnaissance orbiter and dual-frequency SAR (DFSAR) flying on Chandrayaan-2 are analyzed in this article, to constrain the properties of IMPs within some depth from a radar perspective. The SAR datasets are first orthorectified and coregistered with the optical images, which allow for the interpretation of different terrains in small IMPs. Then, the polarimetric and scattering characteristics of different IMP-associated terrains are extracted and analyzed. Results show that the higher mounds in Ina, Sosigenes and Cauchy-5, as well as the lower uneven units in Ina and Sosigenes IMPs might be covered by a layer of rock-poor materials, which is similar to the surrounding maria regolith. This layer can be as thick as $>$ 0.6 m for the higher mounds and lower uneven units in Ina. Importantly, radar-bright IMPs are found, whose scattering characteristics are similar to these of the rocky units in Ina and the ejecta of degraded craters. Based on the radar characteristics of IMPs, their possible origins and evolution processes are discussed.

Software composition analysis (SCA) is essential for understanding and optimizing complex C programs, ensuring system reliability and efficiency. Analyzing programs at the binary level provides insights into behavior, performance, and security. However, comprehensive evaluations of both academic and commercialized SCA tools are lacking. To this end, this paper presents a comprehensive evaluation of software composition analysis techniques for accurately identifying components in C/C++ binary programs. The study examines different analysis techniques in terms of accuracy, performance, domain-specific capabilities, and additional abilities such as detecting security vulnerabilities and code reuse potential. The results show that SCA tools reach over 70% accuracy in detecting general libraries and the accuracy drops to less than 45% for libraries in domain-specific software. Commercialized tools exhibit better efficiency and practicalness than academic tools. The evaluation provides insights into the strengths and limitations of various approaches, offering suggestions for SCA development and the selection of the most suitable tools.

Distributed Optical Fiber Sensing Technology, with advantages such as long detection distance, resistance to electromagnetic interference, and easy maintenance, is widely applied in various fields including security monitoring, electrical safety, and aerospace. However, the current Distributed Optical Fiber Sensing Technology faces challenges such as short localization distance, limited range for stress measurement, low accuracy, and time-consuming processes. This study proposes a novel approach for extracting Brillouin frequency shift signals using an adaptive gradient descent algorithm (Adam Algorithm). A Brillouin strain testing system based on heterodyne coherent detection is also constructed. Experimental results show that the distributed strain testing system using the Adam Algorithm can achieve accurate and fast measurement of maximum strain up to 9500 µϵ within a range of 10 kilometers. The average strain measurement deviation is 32.88 µϵ, and the time required for frequency shift extraction is less than 18.5 ms. This method provides a theoretical and experimental basis for the application of BOTDR Distributed Optical Fiber Sensing Technology.

With the growing public attention to data privacy protection, the problem of data silos has been exacerbated, which makes it more difficult for crowd intelligence technologies to get off the ground. Meanwhile, Federated Learning (FL) has received great attention for its ability to break data silos and jointly build machine learning models. To crack the data silo problem in crowd intelligence, we propose a new Crowd Federated Learning (CFL) framework, which is a two-tier architecture consisting of a cloud server, model owners, and data collectors, that enables collaborative model training among individuals without the need for raw data interaction. However, existing work struggles to simultaneously ensure the balance of incentives among data collectors, model owners, and cloud server, which can affect the willingness of sharing and collaboration among subjects. To solve the above problem, we propose a hierarchical incentive mechanism named FedCom, i.e., Crowd Federated Learning for Contribution matching, to match participants’ contributions with rewards. We theoretically prove that FedCom has contribution matching fairness, and conduct extensive comparative experiments with five baselines on one simulated dataset and four real-world datasets. Experimental results validate that FedCom is able to reduce the computation time of contribution evaluation by about 8 times and improve the global model performance by about 2% while ensuring fairness.

As a new integrity authentication technology, subject-sensitive hashing has the ability to achieve subject-sensitive authentication for high-resolution remote sensing (HRRS) images and can provide a security guarantee for their subsequent use. However, existing research on subject-sensitive hashing focuses on improving the structure of the deep neural network of the algorithm to improve the algorithm's performance, which makes it necessary to reconstruct the training dataset or modify the network structure in the face of different integrity authentication requirements. In this article, we delve into the impact of dropout on subject-sensitive hashing and propose a stepwise-drop mechanism to address the robustness and tampering-sensitivity requirements of subject-sensitive hashing. On this basis, a network named stepwise-drop and transformer-based U-net (SDTU-net) is proposed for subject-sensitive hashing of HRRS images. SDTU-net can use our proposed stepwise-drop mechanism to determine the drop rate of different network layers, which makes it possible to adjust the algorithm performance without changing network structure and training data. Experiments show that our SDTU-net based subject-sensitive hashing has better overall performance compared with existing algorithms, especially at medium and low thresholds. Our approach solves the problem that the existing algorithms cannot balance robustness and tamper sensitivity at low thresholds.

Open-source software has revolutionized the field of software development, providing a collaborative and transparent approach that encourages knowledge sharing and innovation. However, the adoption of open-source software in automotive systems introduces security concerns that require careful scrutiny and management. While previous studies have investigated the general open-source software ecosystem, this paper focuses specifically on the unique characteristics and challenges of open-source software in automotive vehicles. To achieve these goals, we examine real-world examples and case studies to analyze the impact of open-source software on automotive systems. We extracted and collected 4092 open source components from real-world automotive firmware and compared the differences and similarities to 20010 commonly used components in operating systems such as Linux. The results show that the automotive firmware contains 79.8 open source components on average, which account for 15.84% of all binary files in the firmware. Moreover, 61.15% detected libraries are automotive-specific and are different from the commonly used components. By gaining insights this research provides valuable recommendations to improve automotive security by building the database of automotive-specific open source libraries.

Polarimetric synthetic aperture radar (SAR) provides an effective way to understand the physical properties of planetary surfaces. Rocks are the products of specific geological events, recording the evolution of impact craters on airless bodies. Multiple-scattered radar echoes from the surface or near-surface rocks generally exhibit high circular polarization ratio (CPR). In this article, we systematically investigate the relationship between radar CPR and rock abundance of lunar regolith using miniature radio frequency (Mini-RF) SAR data. A total of 51 craters with diameters of 11.8–200 km are selected for analysis. The mean values of Mini-RF CPR and Diviner rock abundance are calculated for the interior and exterior of each crater. Statistical results show a close correlation between Mini-RF CPR and Diviner rock abundance, where high radar CPR and Diviner rock abundance are observed at the blocky ejecta. However, we find the difference of mean CPR value between crater interior and exterior is smaller than that of Diviner rock abundance. This may be attributed to the near-surface rocks that contribute to radar CPR but are not considered in the inversion of Diviner rock abundance. Therefore, the concept of volumetric rock abundance is introduced, indicating the proportion of rocks with diameters larger than one-tenth of the radar wavelength within the detection area. A three-component CPR model is then developed to link radar CPR to volumetric rock abundance. Furthermore, an inverse power-law relationship between volumetric rock abundance and crater ages is discovered, which further indicates the feasibility of using SAR data to estimate rock abundance of lunar regolith.

Accurate power load forecasting is crucial to the safe and stable operation of power systems. In the context of spot market, the dynamically changing real-time market tariff gives the “commodity” property of electricity and changes the electricity consumption behavior of customers and the electricity consumption at each time, which significantly aggravates the difficulty of electricity load forecasting. To address the problems of many influencing factors, difficult input sequence selection and insufficient feature extraction capability of the prediction model, we propose a hybrid prediction model that combines novel input sequence selection and longitudinal crossover algorithm (CSO) to optimize deep belief networks. Firstly, the real-time electricity price is incorporated as an influencing factor into the prediction model input, which improves the prediction model influencing factor system in the market environment; secondly, the similarity between the historical load sequence influencing factor and the sequence of load influencing factors of the day to be predicted is considered from two perspectives of distance and trend, and the historical load sequence is reasonably selected using the comprehensive similarity, and then the prediction model input sequence is determined; finally, the deep belief Finally, the load forecasting model is constructed by using deep belief network, and the key parameters such as threshold value of the forecasting model are optimized by CSO algorithm to achieve accurate power load forecasting. The proposed hybrid forecasting model is validated by comparing and analyzing the forecasting results of different input sequences and different methods with the simulation of Singapore electricity market data. In addition, input series of different time scales are set to find the best sample set to avoid the reduction of forecasting accuracy caused by data redundancy in the sample set, and a sample set of 6 months is appropriate for load forecasting by the method in this paper.

Cognitive impairment is typically reflected in the time and frequency variations of electroencephalography (EEG). Integrating time-domain and frequency-domain analysis methods is essential to better understand and assess cognitive ability. Timely identification of cognitive levels in early Parkinson’s disease (ePD) patients can help mitigate the risk of future dementia. For the investigation of the brain activity and states related to cognitive levels, this study recruited forty ePD patients for EEG microstate analysis, including 13 with mild cognitive impairment (MCI) and 27 without MCI (control group). To determine the specific frequency band on which the microstate analysis relies, a deep learning framework was employed to discern the frequency dependence of the cognitive level in ePD patients. The input to the convolutional neural network consisted of the power spectral density of multi-channel multi-point EEG signals. The visualization technique of gradient-weighted class activation mapping was utilized to extract the optimal frequency band for identifying MCI samples. Within this frequency band, microstate analysis was conducted and correlated with the Montreal Cognitive Assessment (MoCA) Scale. The deep neural network revealed significant differences in the 1-11.5Hz spectrum of the ePD-MCI group compared to the control group. In this characteristic frequency band, ePD-MCI patients exhibited a pattern of global microstate disorder. The coverage rate and occurrence frequency of microstate A and D increased significantly and were both negatively correlated with the MoCA scale. Meanwhile, the coverage, frequency and duration of microstate C decreased significantly and were positively correlated with the MoCA scale. Our work unveils abnormal microstate characteristics in ePD-MCI based on time-frequency fusion, enhancing our understanding of cognitively related brain dynamics and providing electrophysiological markers for ePD-MCI recognition.

Internal discharges in composite insulators often cause abnormal heating, interface failure and erosion of the core rod. However, the mechanism and the exact process of the degradation of HTV shed and FRP core rod caused by internal discharges are still unclear. In this work, abnormal heating insulator samples are taken from an AC 500 kV extra-high voltage (EHV) transmission line. Macro- and micro-scopic morphological observation, hardness tests, industrial CT tests and chemical property tests are applied on the samples. The cause of heating and the deterioration mechanism of each part of insulators is further explored. Cavities and air holes are found in the HTV skirts in this batch of insulators, which are generated during manufacturing. These cavities lead to the distortion of electrical field and become the origin of the discharges. Decomposition of polydimethylsiloxane leads to the hardening and chalking of the HTV shed. It is also found that the adhesive in the rod-housing interface fails during operation. Epoxy resin in core rod decomposes in a multi-factor aging environment, leading to the decay-like aging of the core rod surface. As the aging progress develops inward, the insulator will eventually lose its mechanical properties. It is inferred that the internal discharge is the main reason of heating and deterioration of composite insulators. Moreover, long-term discharge will lead to the decay-like aging of core rods.

Land subsidence is a common geological hazard in urban areas. It can be caused by human activities such as land reclamation and subway construction. However, rapid urbanization can accelerate the progress of these activities. Therefore, spatial-temporal analysis of land subsidence can provide a guarantee for rapid urbanization progress. In this study, Shenzhen, one of the fastest urbanizing cities in China, is selected for continued land subsidence monitoring. We collected 153 Sentinel-1A images from December 2015 to July 2021 and utilized an interferometric point target analysis method to generate the land subsidence in Shenzhen. Additionally, we analyzed the relationship between land subsidence and typical urbanization progress: land reclamation, transportation network, and urban construction. The results show that land subsidence mostly occurs in the western reclamation area, with a maximum rate of −74.9 mm/a in Qianhai Bay. Areas with larger quaternary marine sediment thickness in the range of 10–20 m are more likely to experience subsidence. Obvious subsidence (

DNS root servers are located at the top of the domain name system and are the cornerstone of the Internet. Currently, root servers deploy numerous root instances using anycast technology. Introducing root instances can improve the parsing performance of root servers and the user access experience. However, we found that some root instances do not show optimal performance, and users cannot access the closest root instance when accessing the root instance or even cross-border access. This paper deploys three types of operator detection points in 31 provincial-level administrative regions in mainland China. Each detection point requests the NS record of the top-level domain name from the root instance server introduced in China to obtain the access data of the root instance server hit by domestic users. At the same time, we propose two methods to determine whether users have achieved the optimal choice of root instance, including the method based on the shortest AS path and the method based on geographical distance. In these two methods, we analyze the optimal selection of root instances for each root server. Finally, we analyze the cross-border access of users and find that China Telecom users are more likely to access the root instance across borders.

Ultra-short pulses with high repetition frequency have great application prospects in the field of nano-optics. Here, in the case of continuous wave incidence, the femtosecond pulses with THz repetition frequency are achieved in the transmission system consisting of quantum emitters (QEs) and plasmonic resonators. The generation mechanism of the ultra-short pulses with high repetition frequency is elucidated by semi-classical model. Attribute to the presence of the two-level QEs, the field amplitude in plasmonic resonator is oscillating with time, resulting in the transmittance of the system behave as the form of pulse oscillation. Moreover, the pulse repetition frequency and extinction ratio can be freely controlled by the incident light intensity and QEs number density to obtain the required ultra-short pulses at nanoscale. This also has potential applications in high-speed signal processing fields and generating optical clock signals that can be used in optical computing.

Deep learning methods have shown great promise in automatically extracting features from hyperspectral images (HSIs) for classification purposes. Recently, researchers have recognized the importance of high-order feature interactions—capturing relationships between features in different image regions—in extracting discriminative features. Despite their effectiveness, the existing deep learning models for HSI classification often overlook high-order feature interactions, resulting in suboptimal performance. To address this issue, we propose a novel spectral–spatial multiorder interaction network (S$^{2}$MoINet) for HSI classification. The proposed framework can effectively extract highly discriminative features by leveraging correlations between features in different locations, significantly improving the classification accuracy. More specifically, we design a multiorder spectral–spatial interaction block in the framework to extract the high-order and generalized features by leveraging the interaction between spatial and spectral features. Based on experimental results from four public HSI datasets, it has been shown that the proposed S$^{2}$MoINet delivers optimal classification results when compared to other state-of-the-art methods.

Depression is a severe psychiatric illness that causes emotional and cognitive impairment and has a considerable impact on patients’ thoughts, behaviors, feelings and well-being. Moreover, methods for recognizing and treating depression are lacking in clinical practice. Electroencephalogram (EEG) signals, which objectively reflect the internal workings of the brain, is a promising and objective tool for recognizing and diagnosing of depression and enhancing clinical effects. However, previous EEG feature extraction methods have not performed well when exploring the intrinsic characteristics of highly complex and nonstationary EEG signals. To address this issue, we propose a regularization parameter-based improved intrinsic feature extraction method of EEG signals via empirical mode decomposition (EMD), which mines the intrinsic patterns in EEG signals, for depression recognition. Furthermore, our method can effectively solve the problem that EMD fails to extract intrinsic features. In this method, we first select an appropriate regularization parameter to generate the regularization matrix. Next, we calculate the sum of the matrix products of the IMFs and the regularization matrix and leverage the inverse of this matrix to extract the intrinsic features. The classification results of our method on four EEG datasets reached 0.8750, 0.8850, 0.8485 and 0.7768, respectively. In addition, compared with the iEMD method, our method requires less computational costs. These results support our claim that our method can effectively strengthen the depression recognition performance, and our method outperforms state-of-the-art feature extraction approaches.

Multimodal transfer learning aims to transform pretrained representations of diverse modalities into a common domain space for effective multimodal fusion. However, conventional systems are typically built on the assumption that all modalities exist, and the lack of modalities always leads to poor inference performance. Furthermore, extracting pretrained embeddings for all modalities is computationally inefficient for inference. In this work, to achieve high efficiency-performance multimodal transfer learning, we propose VideoAdviser, a video knowledge distillation method to transfer multimodal knowledge of video-enhanced prompts from a multimodal fundamental model (teacher) to a specific modal fundamental model (student). With an intuition that the best learning performance comes with professional advisers and smart students, we use a CLIP-based teacher model to provide expressive multimodal knowledge supervision signals to a RoBERTa-based student model via optimizing a step-distillation objective loss—first step: the teacher distills multimodal knowledge of video-enhanced prompts from classification logits to a regression logit—second step: the multimodal knowledge is distilled from the regression logit of the teacher to the student. We evaluate our method in two challenging multimodal tasks: video-level sentiment analysis (MOSI and MOSEI datasets) and audio-visual retrieval (VEGAS dataset). The student (requiring only the text modality as input) achieves an MAE score improvement of up to 12.3% for MOSI and MOSEI. Our method further enhances the state-of-the-art method by 3.4% mAP score for VEGAS without additional computations for inference. These results suggest the strengths of our method for achieving high efficiency-performance multimodal transfer learning.

The grid-connection of large-scale and high-penetration wind power poses challenges to the friendly dispatching control of the power system. To coordinate the complicated optimal dispatching and rapid real-time control, this paper proposes a hierarchical cluster coordination control (HCCC) strategy based on model predictive control (MPC) technique. Considering the time-varying characteristics of wind power generation, the proposed HCCC strategy constructs an improved multi-time-scale active power dispatching model, which consists of five parts: formulation of cluster dispatching plan, rolling modification of intra-cluster plan, optimization allocation of wind farm (WF), grouping coordinated control of wind turbine group (WTG), and real-time adjustment of single-machine power. The time resolutions are sequentially given as 1 hour, 30 min, 15 min, 5 min, and 1 min. In addition, a combined predictive model based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), wavelet thresholding (WT), and least squares support vector machine (LSSVM) is established. The fast predictive feature of this model cooperates with the HCCC strategy that effectively improves the predictive control precision. Simulation results show that the proposed HCCC strategy enables rapid response to active power control (APC), and significantly improves dispatching control accuracy and wind power accommodation capabilities.

With the construction of the modern power system, power load forecasting is significant to keep the electric Internet of Things in operation. However, it usually needs to collect massive power load data on the server and may face the problem of privacy leakage of raw data. Federated learning can enhance the privacy of the raw power load data of clients by frequently transmitting model updates. Concerning the increasing communication burden of resource-heterogeneous clients resulting from frequent communication with the server, a communication-efficient federated learning algorithm based on Compressed Model Updates and Lazy uploAd (CMULA-FL) was proposed to reduce the communication cost. CMULA-FL also integrates the error compensation strategy to improve the model utility. First, the compression operator is used to compress the transmitted model updates, of which large norms are uploaded to reduce the communication cost of each epoch and transmission frequency. Second, by measuring the error of compression and lazy upload, the error is accumulated to the next epoch to improve the model utility. Finally, based on simulation experiments on the benchmark power load data, the results show that the communication cost decreases at least 60% with controlled loss of model prediction compared with baseline.

Existing target detection models are large and have multiple network parameters, which can severely slow down the detection speed when deployed on small, low-cost GPU-free Industrial Personal Computers (IPC). As a result, this study proposes a lightweight real-time tomato detection and point-picking integrated network model based on YOLOv5 (TDPPL-Net). Firstly, the algorithm replaces the YOLOv5 backbone with a four-group lightweight downsampling model consisting of Ghost Conv and Ghost Bottleneck to reduce the model size, while adding the attention mechanism SimAM module to improve detection accuracy after each scale’s feature map. Secondly, the Spatial Pyramid Pooling-Fast (SPPF) network structure is used and the convolutional layers in the (Feature Pyramid Network and Path Aggregation Network) FPN+PAN structure are replaced with a depth-separable convolution to reduce the computational effort. Finally, the center of the bounding box is used as the picking point, and the corresponding depth information is obtained in combination with the Intel RealSense D435 camera, which is converted into 3D coordinates under the robot arm coordinate system after hand-eye calibration. The experimental results show that TDPPL-Net reduces the number of parameters by 59.84% compared with the original YOLOv5, the model volume is only 40% of the original, the mAP is 93.36, and the real-time detection speed on the IPC without GPU acceleration is 31.41 FPS, which is 170.31% higher than YOLOv5. The TDPPL-Net increases detection speed on low-performance equipment without compromising detection accuracy. It can detect and locate tomato picking points in real-time in the complex natural environment, which can meet the working requirements of harvesting robots.

Modeling collective behavior is a way to better understand the mechanisms that govern collective animal behaviors. Traditional rule-based modeling methods rely heavily on human prior knowledge and may not provide a proper explanation of the phenomenon of collective behaviors. This paper proposes a Deep Q-Networks (DQN)-based modeling method for fish school. Firstly, an individual’s state (continuous value) is expressed by the angle between its direction and the average direction of its perceived neighbors. An individual’s action is represented with discretized turning angle. Then, the reward function is constructed with the change in the number of neighbors. And finally, the neural network structure is constructed to represent the Q-value function and is trained by the DQN algorithm. The proposed approach is tested in two scenarios: single-learner and multi-learner. Results show that in both scenarios the proposed method can gradually converge and finally obtain a model that can produce collective behavior. On this basis, this paper also deeply analyzes the learned model from the perspectives of average order parameter and collective behavior patterns. It verifies that the behavior pattern generated by the learned model is a highly ordered collective behavior. In addition, we make a comparison between our proposed approach and the Q-Learning algorithm. The results show that our approach not only has a stronger ability to express policy and is better at handling continuous states but also has a more stable learning performance in training.

In order to improve the fault-tolerant control (FTC) ability of the current sensor of the primary permanent magnet linear motor (PPMLM) drive system, the direct thrust force control (DTFC) strategy with single DC-link current sensor is proposed. According to the principle of voltage vector equivalence, the present basic voltage vector is equivalent to two adjacent sub-basic voltage vectors to achieve the equivalence of the present candidate voltage vector. The principle of different phases based on the fact that the DC-link current is equal to or opposite to one of the phase currents is introduced. The output sequence of the two sub-basic voltage vectors is determined by the principle of different phase mode to ensure that the measured currents in the two adjacent periods are any two-phase currents in the three phases, and then the third-phase current is calculated. Finally, the effectiveness and correctness of the proposed DTFC (P-DTFC) is verified by simulation and experiment.

In space-borne synthetic aperture radar (SAR), the sliding spotlight mode can acquire images with both high-resolution and wide-swath in azimuth direction. Due to the significant two-dimensional spatial variance of Doppler parameters, the traditional imaging algorithms based on the conventional range models is not available. In this article, the strategy of continuously varying pulse repetition interval (CVPRI) is likely to lead to a novel approach to dealing with the azimuth variance problem to realize high-resolution wide-swath (HRWS) imaging in azimuth direction for sliding spotlight SAR. First, the eighth-order Taylor expansion of the modified equivalent squint range model (MESRM-TE8) is adopted, and the accuracy of MESRM-TE8 is accordingly explained. Then, the properties of the spatial variance for the MESRM-TE8 are analyzed in detail, based on which the strategy of CVPRI is given theoretically to eliminate the azimuth variance. An improved imaging algorithm based on CVPRI is subsequently proposed to address the azimuthal-variant Doppler parameters and realize a batch data processing of a large scene in azimuth frequency domain. The extended scaling method is integrated in this algorithm to uniformly compensate cubic phase modulation introduced by CVPRI and circumvent azimuth time folding caused by subaperture processing in the focused image. Finally, the effectiveness of the CVPRI strategy and the proposed algorithm is demonstrated by the simulation results.

The technology of retrieving sea surface wind field from spaceborne synthetic aperture radar (SAR) is increasingly mature. However, the retrieval of the sea surface wind field related to the precipitation effect is still facing challenges, especially the strong precipitation related to extreme weather such as tropical cyclone will cause the wind speed retrieval error to exceed 10 m/s. Semantic segmentation and weak supervision methods have been used for SAR rainfall recognition, but rainfall segmentation is not accurate enough to support the correction of wind field retrieval. In this article, we propose to use deep learning to classify the rainfall grades in SAR images, and combine the rainfall correction model to improve the retrieval accuracy of sea surface wind speed. To overcome the challenge of limited training samples, the transfer learning method in fine-tune is adopted. Preliminary results demonstrate the effectiveness of this deep learning methodology. The model classifies rain and no-rain images with an accuracy of 96.2%, and classifies rainfall intensity grades with an accuracy of 86.2%. The rainfall correction model with SAR rainfall grade identified by convolution neural network reduces the root-mean-square error of retrieved wind speed from 3.83 to 1.76 m/s. The combination of SAR rainfall grade recognition and rainfall correction method improves the retrieval accuracy of SAR wind speed, which can further promote the operational application of SAR wind field.

In large-scale image retrieval, the deep learning-based hashing methods have significantly progressed. However, most of the existing deep hashing methods still have the problems of low feature learning efficiency and weak ranking relationship discrimination. To remedy these problems, a novel Ranking-based Deep Hashing Network (RDHN) is proposed for image retrieval in this paper, which integrates the feature learning module and hash learning module into a unified deep hashing network framework to jointly learn a powerful hash function so that the raw images can be mapped to discrete hash codes with significant discrimination. Specifically, a novel difference convolution is designed based on edge detection operators, and then it is uniquely applied to the first convolutional layer of the convolutional neural network (CNN), which can take advantage of the sensitive characteristics of edge detection operators for edge information to extract richer image edge information. Meanwhile, in hash learning, the ranking metric Mean Average Precision (MAP) is optimized using the idea of scaling, and then a ranking loss function based on MAP is carefully designed to enhance the neighborhood ranking capabilities of the hash codes. Furthermore, to reduce the quantification error, a quantization loss function is also designed. Finally, the ranking loss function is combined with the quantization loss function to form the objective function. The proposed method can generate high-quality discrete hash codes while learning to preserve ranking information, effectively improving retrieval performance. Extensive experimental results on three widely used benchmark datasets demonstrate that the proposed method outperforms the state-of-the-art hashing approaches.

Most multi-object tracking methods have achieved good results in tracking multiple pedestrians with Kalman filter, but their tracking performance in crowded scenes is still poor due to pedestrian avoidance and frequent occlusion. In crowded scenes, the pedestrian trajectory prediction with Kalman filter alone is unreliable. In this paper, a two-dimensional field-of-view avoidance force model (AFM) is proposed to assist the Kalman filter prediction by sensing the avoidance force and then complete pedestrian tracking. In the model, each pedestrian has a two-dimensional field of view to perceive the avoidance force, which determines the next predicted trajectory. In real scenes, pedestrians tend to be more concerned about the surrounding area, so different areas are set to simulate the attention mechanisms of pedestrians in real scenes. In the FairMOT model, AFM is used to optimize the pedestrian state values of Kalman filter prediction and the optimized model is trained on the MOT16 dataset. The experimental results on the MOT20 dataset show that compared with the mainstream tracking model FairMOT, our method respectively improves MOTA by 2.7% and IDF1 by 2.2%. Our method also achieves the good performance on MOT15, MOT16, and MOT17 tracking benchmarks.

A D-shaped fiber-optic surface plasmon resonance (SPR) sensor with MXene (Ti3C2Tx) is proposed as the sensing enhancement material. A spin coating method was used to cover the gold film D-shaped fiber SPR sensor with a layer of MXene. Experimentally, the sensor's sensitivity, full width at half maximum (FWHM), and figure of merit (FOM) are discussed. The effect of various MXene concentrations on D-shaped fiber SPR sensors was investigated. The results demonstrate that as the MXene concentration increases, the sensitivity, FWHM, and FOM value of the D-shaped fiber SPR sensor increase as well. The MXene-based D-shaped fiber SPR sensor exhibits the highest sensitivity, measuring 3143 nm/RIU, and the highest FOM value, 15.26. Compared to the Au-film D-shaped fiber SPR sensor, Sensitivity is increased by 53.6%. The results indicate that MXene could be used to enhance the performance of SPR refractive index sensors.

Free-space optical communication (FSOC) systems with variable data-rates can support different missions with different transmission distances and ensure reliable transmission in periods of unfavorable weather conditions. The temporal diversity coherent combining (TDCC) technique can adjust the data-rate in a very large range using a software-defined manner by taking advantage of the digital coherent receivers able to recover the signal field. In this paper, we investigate the relationships between the computational complexity, optical phase alignment error, combining loss (CL) and data-rate for the TDCC-based variable data-rate coherent optical receiver. We also propose a method to minimize the computational complexity while achieving the expected output optical signal-to-noise ratio (OSNR) and the highest data-rate for an arbitrary input OSNR. Numerical simulations and experiments are carried out to validate the analytical expressions and proposed methods. The results provide an efficient tool and useful guidelines for the design of low computation complexity TDCC-based variable data-rate FSOC systems.

Aiming at the difficulty in identifying subtle AC arcs in aviation cables, this paper proposes an arc fault detection method based on the combination of three-dimensional features and convolutional neural network-long short term memory (CNN-LSTM). Firstly, based on the SAE AS5692A standard, the vibration series test, cutting parallel test, and wet arc trajectory parallel test were respectively conducted and the arc current signals under four types of loads were collected to analyze the arc faults under different incentives. Then, the three-dimensional features of arc current including Hurst exponent, inter-harmonic variance, and wavelet energy entropy (H-I-W) were extracted with an improved algorithm so as to enhance the fault identification capability and overcome the limitation of single-dimensional feature detection. Finally, a grid search algorithm was used to find out the optimal parameters, and a three-dimensional reference input CNN-LSTM neural network was designed to detect arc faults. The experimental results showed that the average detection accuracy of the proposed method for the three AC arc faults respectively reached 98.52%, 99.23%, and 98.51%. The real-time performance of the proposed method was better than the comparison methods, proving the feasibility and effectiveness of the proposed method.

Intersection over union (IoU) has been widely adopted to evaluate and select candidate regions in multi-oriented object detection. Intuitively, overlaps between candidates and multi-oriented ground-truth boxes make more sense when assessing the quality of horizontal candidates. However, the horizontal minimum bounding box (HMBB) of the ground-truth box is generally used for the IoU calculation in practice, bringing about biased results. In this article, we propose a novel Splicing Intersection over Union (SIoU) to provide a more preferable metric for horizontal candidate selection when detecting multi-oriented objects. By computing the intersection between the candidate region and the ground-truth box rather than its HMBB, SIoU provides a better description of how much object information a candidate contains. Furthermore, we introduce two variants of constraints for the center of each candidate to ensure its location focusing on the objects. Candidates whose centers deviate too far from the objects will be penalized. We integrate the constraint with SIoU, denoted as constrained-SIoU, to select horizontal candidates more efficiently. Comparative experiments on two datasets of aerial images, DOTA and HRSC2016, demonstrate the effectiveness of the proposed method.

In this paper, a direct thrust force control (DTFC) strategy is proposed for the half-open winding primary permanent magnet linear motor (PPMLM) drive system in order to improve the dynamic response and performance of flux weakening control. The independent voltage active vector (IVAV) is selected as the candidate voltage vector to suppresses the zero sequence current in the proposed DTFC (P-DTFC). Meanwhile, a new switch table is designed to determine the control vector while the complexity of control method is greatly simplified in P-DTFC. The simulation and experimental results verify that P-DTFC maintains better steady-state and dynamic-state performance with lower complexity compared with traditional DTFC (T-DTFC).

Community structure has a more intuitive, physical meaning than how it has traditionally been perceived; this representation plays an important role in networks and is accordingly becoming widely adapted. Consequently, community detection has attracted increasing attention. Although scholars have proposed many community detection methods from different perspectives, due to the complexity, diversity and dynamic characteristics of networks, efficient community detection in many real networks remains a challenge. Inspired by fish school effect in real life, this paper envisions networks as an ecosystem and proposes a novel dynamic model that aims to reveal the communities in a more intuitive way. Relying on the new model, we design a community detection algorithm, known as community detection based on fish school effect (CDFSE). CDFSE has plentiful desirable properties: high-quality community detection, parameters free and notable scalability. To evaluate the performance of CDFSE, this paper employs two widely utilized evaluation metrics and twelve representative algorithms to test the effectiveness of the algorithm in both synthetic and real-world networks. The experimental results show that in most cases, CDFSE is superior to the comparison methods in terms of the quality of community detection.

Video-level sentiment analysis is a challenging task and requires systems to obtain discriminative multimodal representations that can capture difference in sentiments across various modalities. However, due to diverse distributions of various modalities and the unified multimodal labels are not always adaptable to unimodal learning, the distance difference between unimodal representations increases, and prevents systems from learning discriminative multimodal representations. In this paper, to obtain more discriminative multimodal representations that can further improve systems’ performance, we propose a VAE-based adversarial multimodal domain transfer (VAE-AMDT) and jointly train it with a multi-attention module to reduce the distance difference between unimodal representations. We first perform variational autoencoder (VAE) to make visual, linguistic and acoustic representations follow a common distribution, and then introduce adversarial training to transfer all unimodal representations to a joint embedding space. As a result, we fuse various modalities on this joint embedding space via the multi-attention module, which consists of self-attention, cross-attention and triple-attention for highlighting important sentimental representations over time and modality. Our method improves F1-score of the state-of-the-art by 3.6% on MOSI and 2.9% on MOSEI datasets, and prove its efficacy in obtaining discriminative multimodal representations for video-level sentiment analysis.

It has always been an important and arduous task to detect the complex dynamic traffic environment, especially for unmanned driving. Although the existing advanced detection models have reached the speed requirements for detection, the detection accuracy needs to be further elevated to improve the unmanned driving’s safety. How to balance the accuracy and speed of detecting the complex dynamic traffic environment is still the primary problem to be solved for unmanned vehicles. Therefore, this article proposes a detection model of the complex dynamic traffic environment for unmanned vehicles by following the framework idea of YOLOv3. Firstly, we regard MobileNetv3 as the backbone and replace the traditional convolution with the depthwise separable convolution in the whole model to reduce the number of parameters and calculations. Secondly, in enhanced feature fusion layers, we perform the multi-scale fusion of four feature maps by the compress-and-expand module, the SPP module, and the cross-layer bidirectional module of feature fusion to improve the locating accuracy and reduce false detections. Thirdly, we add an IoU loss to improve the accuracy of model regression. Then, we employ the improved clustering algorithm to re-cluster anchor boxes, reducing the time overhead while improving the clustering accuracy. Finally, we compare the proposed model with other advanced detection models in the processed BDD dataset and the KITTI dataset. We verify that the mAP of the proposed model improves notably without loss of detection speed, the number of parameters and calculations decreases dramatically, and the proposed model exhibits a more superior performance.

Optical tunneling is an effect extremely sensitive to wavelength changes. In this article, we use this effect to design a new type of biosensor. The tunneling effect is excited by long range surface plasmon resonance, the energy is concentrated in the sensing medium in the form of standing waves, which lead a very high wavelength sensitivity and quality factor. By using COMSOL for finite element analysis, for the sensor structure of BK7/Cytop /Al2O3/Ag/Al2O3/sensing media/air, a quality factor of 10296 RIU-1 is achieved under angle modulation, a sensitivity of 170,000 nm/ RIU and a quality factor of 11333 RIU-1 are obtained under wavelength modulation. At the same time, the advantage of this structure is that there is no excessive requirement on the refractive index of the material, and a good sensing effect can be achieved even if the material is cheap and easy to prepare. After optimization, this structure is suitable for sensing media with a refractive index of 1.335-1.375, which can be applied to most liquid solutions.

In this work, we fabricated and characterized 4 × 4 parallel flip-chip AlGaN-based micro-LED arrays with varied mesa diameters of 120 µm, 100 µm, 80 µm, and 60 µm. The reported micro-LED arrays have a maximum bandwidth of 380 MHz and a peak wavelength of ∼275 nm. It is found that the electrical and optical characteristics of AlGaN-based micro-LED arrays show strong size dependence for ultraviolet communication (UVC). The differential resistance increases from 28.8 Ω to 112 Ω, the external quantum efficiency (EQE) is increased by ∼30%, and the bandwidth doubles as the diameter of individual micro-LED decreases from 120 µm to 60 µm. Our research proves that tailoring the mesa size of parallel flip-chip AlGaN-based micro-LED arrays can further enhance its bandwidth and promote its application in UVC.

The neural machine translation (NMT) model is a data hungry and domain-sensitive model but it is almost impossible to obtain a large number of labeled data for training it. This requires the use of domain transfer strategy. In order to solve the problem of domain data mismatch, this paper proposes a neural machine translation transfer model based on domain mutual guidance and establishes the continuous impact through the framework of mutual guidance. At the same time, self-ensemble and self-knowledge-distillation are used in these independent domains so that the model will not deviate from the domain too much. Furthermore, the model can better train the models from the batching way of domain data. It mainly uses the pretraining model out of domain, distillation of existing models in domain and data selection in the training process to guide the in-domain model. These are unified in the training framework, so that model training can be continuously and effectively guided in and out of domain. In this study, three typical experiment scenarios were comprehensive tested and our model was compared with many conventional classic methods. The experiment results showed that our proposed “inter-domain transfer training” and “curriculum scheduling agent” was effective and robust. The most important results and findings are that this comprehensive guided training framework (intra-domain and inter-domain) is suitable for the domain transfer in different scenarios, and this framework doesn’t increase the decoding cost.

Towards the massive amount of data generated in our daily work and life, embedded systems, with economical but powerful storage and computing resources, are inevitably becoming the most suitable platform for the Edge Computing for the Internet of Things. However, embedded system servers may also threaten individuals by storing individuals’ private data for years. This paper proposes a Resilient Tag-based Privacy Protection (RTPP) scheme for embedded systems. Specifically, to protect the privacy against the hackers and other non-users, we employ a pseudo-random number encryption technique with the chaos-based principle so that the third party cannot easily steal the private data and reduce the risk of personal privacy leakage. To protect the individuals’ interests, we propose a new approach to controlling the life cycle table of data to enable individuals themselves the flexibility to control the life cycle of private data. Unlike existing data lifetime management methods, the RTPP can support the retrieval of tags in the data life cycle table to control the corresponding privacy while automatically adding or removing tags. Our system automatically adjusted the survival period of private data in the life cycle table through the change of leaf weights, controlled the charge movement on the surface of flash memory, and finally achieved the resilient adjustment process of the life cycle of private data in the embedded system. The security proof and performance evaluation show that the proposed RTPP scheme is provable secure in the automatic privacy lifecycle tuning model for embedded systems and efficient in practice.

Understanding the spatial variability of soil texture classes is essential for agricultural management and environment sustainability. Sentinel-2 data offer valuable vegetation information as proxies for soil properties inference. However, the applications of them in soil texture classification are still limited. This study investigated the usefulness of Sentinel-2 data for predicting soil texture class using an interpretable machine learning (ML) strategy. Specifically, multitemporal Sentinel-2 images were used to get exhaustive vegetation cover information. Basic digital elevation map (DEM) derivatives and stratum were extracted. Three support vector machines with different input parameters (purely DEM derivatives and stratum, purely Sentinel-2, and Sentinel-2 plus DEM derivatives and stratum) were developed. Moreover, in order to improve the transparency in black box ML models, the novel SHapley Additive exPlanations (SHAP) method was applied to interpret the outputs and analyze the importance of individual variables. Results showed that the model with all variables provided desirable performance with overall accuracy of 0.8435, F1-score of 0.835, kappa statistic of 0.7642, precision of 0.8388, recall of 0.8355, and area under the curve of 0.9451. The model with purely Sentinel-2 data performed much better than that with solely DEM derivatives and stratum. The contributions of Sentinel-2 data to explain soil texture class variability were about 17%, 41%, and 28% for sandy, loamy and clayey soils, respectively. The SHAP method visualized the decision process of ML and indicated that elevation, stratum, and red-edge factors were critical variables for predicting soil texture classes. This study offered much-needed insights into the applications of Sentinel-2 data in digital soil mapping and ML-assisted tasks.