Communications, Signal Processing, and Systems

Two important tasks in crowd analysis are crowd counting and crowd localization. In this paper, we introduce map-based crowd counting and localization methods, including density map-based methods, dot mask map-based methods, and distance transformation map-based methods. In addition, we combine the map-based methods with different losses. Finally, we compare the counting and localization performance of map-based crowd counting and localization methods on two benchmark datasets to evaluate the effectiveness of existing maps and loss functions.

With the rapid development of information and communication technology, air-to-ground communication, represented by unmanned aerial vehicles (UAVs), is playing an increasingly important role. UAVs can provide wireless access to devices in the absence of ground network infrastructure. Simultaneously, in order to address the issue of network congestion, the establishment of Device-to-Device Pair (DP) communication within cellular networks has become a way to enhance wireless network capacity. In this paper, we focus on the assignment of scarce spectrum resource and terminal energy consumption, proposing an air-ground communication model and emphasizing the optimization of Energy Efficiency (EE).

In this paper, resource allocation and RIS regulation in reconfigurable intelligent surface (RIS) enhanced orthogonal frequency division multiple access (OFDMA) systems in cloud radio access networks (C-RAN) are investigated. Specifically, we consider the uplink, where the remote radio head (RRH) is compressed by independent quantization of the received signal and has the quantized bits transmitted to the baseband unit (BBU) over a limited capacity fronthaul link. To maximize the total rate of the system, we propose a multi-agent deep reinforcement learning (MADRL) approach. We use the multi-agent double deep Q network (MADDQN) algorithm to optimize RRH selection and subcarrier allocation, and the multi-agent depth deterministic strategy gradient (MADDPG) algorithm to optimize power allocation and RIS reflection coefficient regulation. Simulation results show that the proposed method can efficiently maximize the sum rate of RIS-aided OFDMA uplink while satisfying the constraint of limited fronthaul capacity in C-RAN.

This paper proposes a composite preamble structure based on the fusion of the original Ultra Wide Band (UWB) preamble structure and the Nonlinear Frequency Modulated (NLFM) pulse signal to improve the range-sensing ability of the UWB signal. We focus on improving the range-sensing performance of UWB signals with composite preamble structure. The short-time NLFM signal is utilized in the composite preamble structure to locate the target, so as to lower the target misjudgment rate. Simulation results show that the UWB signal with a composite preamble structure improves the target detection accuracy and reduces probability of target misjudgment, outperforming the original UWB signal.

In the task of image inpainting, it is common to utilize a CNN-based encoder-decoder architecture to extract the feature information from the damaged image, achieving satisfactory restoration results. However, these methods often struggle to achieve high-quality restoration for images with varying degrees of damage. In this paper, propose a two-stage inpainting model. Firstly, leverage the powerful contextual capturing capabilities of the Transformer to form a coarse recovery network, so as to roughly fill holes of different sizes. Secondly, employ a predicted filtering kernel network to perform fine restoration, building upon the coarse restoration. Method conducted qualitative and quantitative experiments on the CelebA and Places2 datasets, demonstrating the superiority of our proposed method.

The problem of restoring high-resolution images from blurry images has long been a concern, and traditional methods of directly interpolating low-resolution images to obtain high-resolution images are simple but ineffective. Inspired by SR3, we propose a super-resolution model of human faces based on the diffusion model, which achieves super-resolution through a random iterative denoising process. In this paper, we have used a residual block that integrates multi-scale spatial attention and coordinate attention. Additionally, we have enhanced the restoration of image details through a global attention model. These advancements effectively address the discrepancy between automated evaluation metrics and human perception in high-frequency details for super-resolution models. Through evaluation of the standard eight-fold super-resolution task on CelebA-HQ, our model performs well and achieves competitive scores on SSIM and PSNR metrics.

Resource management in 5G/WiFi coexisting millimeter wave networks is the key to support the high throughput demand in hotspots. A joint multi-dimensional resource allocation algorithm based on block coordinate descent is proposed in this paper, where the original throughput maximization problem is divided into three sub-problems, i.e., user association, beamforming design as well as spectrum allocation, and solved iteratively. Specifically, the method of matching exchange and generalized Rayleigh quotient are used for optimizing user association and digital beamforming, respectively. The simulation verifies that the algorithm achieves better performance in terms of the uplink throughput in 5G/WiFi coexisting networks.

Deep learning makes remarkable progress in the application of remote sensing image processing, particularly in the cloud image segmentation field. The encoder-decoder structure in deep learning is widely employed for cloud image segmentation tasks. The encoder extracts high-level semantic features from the input cloud image, while the decoder restores the semantic features to generate pixel-level segmentation results. Furthermore, skip connections are adopted to connect the encoder and the decoder. In this paper, we introduce and evaluate the representative encoder-decoder struture methods for cloud image segmentation. We focus on the design of encoder, decoder and skip connections. We conduct comparative experiments on cloud image datasets and analyze the encoder-decoder structure with different layers.

The TLD (Tracking-Learning-Detection) algorithm acquires target features through continuous learning from the initial target, enabling prolonged target tracking. However, during the tracking process, variations in lighting conditions, occlusions, and target movements significantly affect its tracking performance. Consequently, the traditional TLD algorithm exhibits high complexity and computational time in its detection module. It also suffers from poor classification performance in low-light environments, inability to meet real-time and accuracy requirements, among others. To address these issues, this study proposes improvements to the detection module of the TLD algorithm. Specifically, we incorporate the LBP (Local Binary Patterns) texture features into the detection module of the TLD algorithm, utilizing them for sample classification to enhance the classification performance. Additionally, we employ the Markov algorithm to predict facial motion, reducing the detection area of the module and improving the algorithm’s real-time capabilities. Experimental results validate that the proposed algorithm demonstrates improved tracking accuracy and real-time performance.

Binocular vision refers to the process of both eyes simultaneously focusing on an object and accurately reflecting the external spatial environment. Binocular vision is constantly required in daily life. With the rapid development of artificial intelligence technology, its application in binocular vision can assist doctors in rapid diagnosis and treatment, which is of great significance for the early detection and treatment of strabismus and amblyopia. This paper primarily researched the problem of convergence and accommodation anomalies in binocular visual function. Firstly, the SMOTE algorithm was used to oversample the dataset and balance the sample of each label category. Then, the random forest algorithm was employed as the basic model for the three training processes in the boosting algorithm. Finally, the models generated from these three trainings were integrated, and the final result was obtained by voting. Comparing the performance differences of the models on the test set before and after the three sampling trainings indicated that the integrated model had a better classification effect on the dataset.

Digital beam forming (DBF) synthetic aperture radar is an important technological approach for achieving high-resolution and wide-swath imaging. However, the unavoidable channel mismatch among received channels is a key issue. Channel imbalance deserves special attention in DBF system since the uncompensated channel errors may cause sidelobes rising, time-variant shift, a broadening of the impulse response, as well as the error of beam-steering direction. Thus, this letter addresses the impact of channel mismatching errors for digital beamforming (DBF) system in elevation and proposes a high accurate calibration method. Simulation experiment shows the effectiveness of the proposed method.

Spring flow prediction is the basis for water resources management, allocation, and effective utilization. To improve the accuracy of spring flow prediction, a hybrid model is used to predict, which combines variational modal decomposition (VMD), long and short-term memory (LSTM) network, and attention mechanism to overcome the endpoint effect and modal confounding problems of traditional empirical modal decomposition. This study explores the performance of VMD-LSTM-Attention and VMD-LSTM, LSTM models in spring water prediction. The experimental results confirm the effectiveness of VMD-LSTM-Attention in spring water prediction. Therefore, this hybrid model is robust and superior for predicting highly non-smooth and non-linear watersheds and can provide a reference for practical hydrological prediction.

The channel coding blind identification technology identifies the channel coding parameter by analyzing the coded data, then realizes the breakthrough from signal layer to information layer, which is widely used in signal monitoring, information acquisition, intelligent communication and other fields. This paper proposes a parity-check matrixbased F-LDPC codes parameter identification method by studying the F-LDPC codes coding mechanism and parity-check matrix generation principle, which can identify the repeat parameter and the puncture parameter of F-LDPC codes according to the parity-check matrix weight characteristics, and recover the equivalent random interleaving of FLDPC codes according to the parity-check matrix structure characteristics. The simulation results show that this method can achieve the recognition of the repeat parameter, the puncture parameter and the random interleaving of regular F-LDPC codes under the condition that the complete parity-check matrix is known.

Large scale distributed neural networks have demonstrated promise for various inference tasks in Internet of Things (IoT) devices, including intelligent security monitoring and defense against network threats. However, the massive amounts of data generated by IoT applications and limited computational capabilities present significant challenges in implementing typical applications, such as secure protocols for data confidentiality. Over-The-Air (OTA) computation, a recently proposed physical layer computing architecture, has great potential to address these issues. In this paper, we propose an OTA distributed neural network with the mutual benefit of joint computing and communication. However, the open channel environment in which the network’s forward computation is implemented renders OTA-based joint computing and communication methods vulnerable to replay attacks, thereby compromising the accuracy of the network performance and wasting valuable bandwidth resources due to backpropagation of contaminated information during OTA computing. A threat model of network is established to investigate the impact of replay attacks during the iterative process. Our analysis and numerical results demonstrate that the replay attacks have a significantly impact on the network. Specifically, the test accuracy rate decreases from 85 to 35%, and the convergence rate decreases by an average of $$40\%$$ 40 % . When the number of iterations is set to 500, the success probability of replay attacks is 0.378.

Spaceborne Synthetic Aperture Radar (SAR) is a powerful remote sensing tool for large scale Earth observation. The combination of Scan-On-Receive (SCORE) mode and digital beamforming (DBF) has been proposed to improve signal-to-noise ratio and attenuate the range ambiguities. Although advantageous, the shape of the narrow receive beam may affect the chirp signal spectrum and introduce pulse extension loss. Considering that the pulse extent on the ground varies from the near range to the far range, this paper proposes a novel real-time operation mode based on optimized sidelobe weighting to the receive channels, which reduces the pulse extension loss and the range ambiguity, particularly at large incidence angle.

The digital filter is an important device for the wireless communication system. This paper first introduces the principles of constrained least squares (CLS) and particle swarm optimization (PSO) algorithms, then gives the design index requirements of the filter and the feasible PSO adaptation value function, and then proposes a specific implementation algorithm for the PSO design filter. Finally, simulations are performed to study the effects of PSO algorithm parameters and the adaptation value function on the filter performance, and better algorithm parameters are derived, and a digital filter is designed accordingly to meet the index requirements.

The Doppler shift estimation has impacts on the performance of vehicular communication systems. This paper analyzes the influence of algorithm parameter on the Doppler shift estimator based on level crossing rate (LCR), in which the level ratio is chosen as the analyzing parameter. Through the simulation with variant level ratios, signal-to-noise ratios (SNR) as well as the Doppler shifts (or vehicle speeds equivalently), this paper comprehensively analyzes the performance variation of Doppler shift estimator, and finds that higher SNRs lead to smaller estimation errors, while an optimal level ratio will further reduce the estimation error. Moreover, this paper also finds that the optimal level ratios are different for different vehicle speeds. Our studies will benefit the practice of LCR-based Doppler shit estimator.

In recent years, the emerging technique of device-free sensing (DFS) has gained popularity for foliage penetration (FOPEN) target recognition. This popularity is primarily attributed to its inherent advantage of not requiring specialized sensing equipment beyond wireless transceivers. Concerning weather variations, DFS heavily relies on labeled data for model training, which necessitates the annotation of samples for each weather environment. However, this annotation process proves impractical for real-world applications, especially under adverse weather conditions. To address this issue, this paper presents an unsupervised domain adaptation (UDA)-based cross-weather FOPEN target recognition system (CW-FTRS). Experimental results validate that the proposed method achieves an average accuracy of over 72% in unseen weather conditions using only unlabeled data samples.

Through research on the issues of data nodes being prone to loss, disorder, and failure in cloud computing data management systems, Designed cloud data recovery technology based on error correction encoding. The design of cloud data recovery technology mainly includes three types: data complete recovery technology, data precise positioning recovery technology, and partial data precise positioning recovery technology. In complete recovery technology, the new data encoding generated through recovery includes the storage point encoding of the disaster recovery system failure, as long as the recovered disaster recovery system supports maximum offset distance encoding. In precise positioning recovery technology, the system needs to accurately recover invalid or lost data encoding. Partial precision positioning recovery technology is a hybrid technology based on complete recovery technology and precision positioning recovery technology, and is not the focus of this article.

This paper focuses on the design of core digital signal processing (DSP) algorithms for optical orthogonal frequency division multiplexing (O-OFDM) systems, specifically digital baseband receivers. The two most critical DSP techniques, synchronization and channel estimation, are discussed. Symbol synchronization and sampling frequency synchronization are two synchronization techniques that are crucial for proper system function. Deviations in symbol synchronization timing points can lead to inter-symbol interference (ISI) and inter-channel interference (ICI), while sampling frequency offset (SFO) can cause amplitude attenuation, phase rotation, and ICI interference of OFDM symbols in the frequency domain. Following comprehensive simulation verification, the optimized DSP algorithm proposed in this paper has been demonstrated to reduce the complexity of FPGA implementation, improve the operational speed of the system, and provide robust support for achieving high-speed O-OFDM systems.

One approach to improve the performance of Low-Density Parity-Check (LDPC) code was to address the presence of short cycles. In this paper, a variable-degree scoring algorithm was presented, designed for both general LDPC codes and Quasi-Cyclic (QC) LDPC codes, with the goal of eliminating short cycles and evaluating the impact on decoding performance. Scores were assigned to the non-zero elements in the LDPC codes based on their involvement in short cycles by the algorithm, and the elements with the highest scores were subsequently modified. Our algorithm effectively eliminated short cycles and improved decoding performance.

In order to reduce the safety hazards of landslide geological hazards to railway operation, a monitoring and early warning system for landslide geological hazards using Zigbee wireless sensor network was designed. The system mainly includes a laser sensor wireless data acquisition terminal layer, a wireless data aggregation layer, a 4G transmission network layer, and a ground monitoring center. The hardware circuit design of the system mainly includes the main control CC2530 circuit connection design, laser ranging sensor circuit connection design, and RS484 communication bus circuit connection design. The system software mainly includes the coordinator software process, terminal node software process, and Modbus RTU data packet format design. System testing shows that when network nodes are deployed within 70 m, the system's packet loss rate can be controlled within 5% for smooth operation.

Multispectral images have rich spatial and spectral information which contain great application superiority. Therefore, effective compression of multispectral images is crucial. This paper proposes an end-to-end network architecture based on prediction networks to complete multispectral image compression tasks. Specifically, the feature extraction module can extract spatial and spectral information effectively and reduce information redundancy. The prediction module is able to predict the original image and obtain the residual one according to the reference spectral image and the extracted features. All modules are jointly optimized by a single loss function. The experimental results show that proposed compression framework outperforms conventional methods, including JPEG2000 and 3D-SPIHT.

In order to solve the problem that the traditional Stationary bicycle cannot collect heart and lung heart rate data and the professional Monark bike is expensive, a two-stage load cardiopulmonary endurance monitoring bike Motor system was designed. The hardware part of the system mainly completes the design of heart rate data measurement and collection module, power load measurement and collection module, and resistance regulation module. The collection of motion data is completed through pressure sensors, torque motors, and angle sensors. The software mainly completes the design of the data collection process for the lower computer and the data analysis and display design for the upper computer APP. The system has designed a two-stage load cardiopulmonary endurance test scheme, which uses the VO2max index to achieve the evaluation and monitoring of the user's cardiopulmonary heart rate level. The test results indicate that the cardiopulmonary endurance monitoring system has a significant correlation with the Monark power cycle, verifying the accuracy of the system.

In order to solve the problem of decentralized management and low intelligence of traditional smart home, a wifi smart home system with centralized management, unified deployment and wireless control is designed. The system realizes the design of hardware system and software system of smart home system, and realizes the real-time collection of home data such as air quality, light, gas, temperature and humidity through various sensors. It also realizes the collection of video and audio data of home environment through wireless network, and completes the remote operation of home appliances through operation instructions. After system testing, the packet loss rate of WiFi wireless data transmission is ≤ 0.4%, and the accuracy rate of speech recognition is 92%. The mobile app achieves remote operation and real-time monitoring of the home environment, meeting the control requirements of the home environment in smart homes, and has certain promotion value.

As one of geologic factors affecting the sustainable development of regional economy and society, the importance of land subsidence is increasing gradually. And then a high-precision prediction of land subsidence trends is of great significance for the disaster prevention and reduction. However, the common prediction methods have shortcomings with strong constraints, difficult parameter model selection, and weak generalization ability. Therefore, a combination neural network based on CNN and LSTM is proposed. Firstly, the SBAS InSAR analysis method was used to obtain time series monitoring results. Then, the multi-dimensional feature extraction of land subsidence was carried out through spatial interpolation and CNN to extract the features in spatial and temporal. Finally, LSTM was used for the feature learning and prediction of land subsidence. And an area near seaside is chosen for the predication experiment, the cumulative subsidence amount and the subsidence changes in different subsidence interval were reflected from the predicted subsidence results. Three accuracy evaluation were carried out to verify the effectiveness of the combination method. The absolute error of the model prediction was less than 4 mm from regional scale, the mean square error of was less than 2.5 mm from point scale, and the average prediction accuracy was improved about 12% through comparing with common method.

In order to solve the shortcomings of the traditional spherical robot, such as less application scenarios, poor practicability and high cost, a spherical mobile robot system with visual object autonomous recognition is designed. The hardware structure of the robot mainly includes high-definition image acquisition bpi-d1, sensor data fusion processing master STM32F103, motor drive, Bluetooth transmission and power management. In order to complete the self-balancing state of the robot in the moving process, the PID control algorithm is designed. In terms of software, the overall functional flow design and self-balancing control flow design of the robot are completed. Through the system software and hardware test, it can be seen that the average maximum deflection angle of the robot's non-interference self-balancing is about 19.86°, the maximum average deflection angle of the disturbed self-balancing is about 23.86°, the average self-balancing time is about 1.78 s, and the disturbed self-balancing time is within 3S, which meets the original design intention of the robot.

In order to solve the problem of whether the environmental noise along the railway exceeds the standard, a noise monitoring system along the railway has been designed that can achieve the integration of noise data acquisition, A/D conversion, Ethernet transmission, algorithm processing, data storage, and interface monitoring. The system uses ADS1256, TMS320VC5509, and ADR4525 chips to complete system signal conditioning, signal conversion, and signal processing, and completes data network transmission to the OneNet cloud through the Ethernet module W5500. The upper computer of the system uses VS.NET and C# to achieve cloud data reading, and the lower computer implements DPS data processing through the CCS6 environment and C program. The system test shows that the data acquisition accuracy is 0.2 level, and the maximum frequency division vibration level and Z vibration level values are used as noise research and judgment indicators to meet the system monitoring purpose.

For the session recommendation scenario, user behaviors in the same session are intrinsically related, and the context information of user session behavior is introduced into the session, and the behavior in the session is modeled. Introducing a dual attention mechanism, assigning different weights to user input behavior data, constructing a model, and conducting experiments on two public datasets. Compared with the benchmark model, this model has improved in various evaluation indices, proving its effectiveness.

Channel knowledge map (CKM) is a promising technology that can build a bridge between the environments and communication strategies for joint sensing and communication. This paper proposes a CKM construction framework involving multi-dimensional channel information in three-dimensional (3D) space including the channel knowledge extraction and completion of sparse channel knowledge. For channel knowledge extraction, the constant false alarm probability (CFAR) method is utilized to extract the effective channel knowledge from the channel impulse response (CIR). On this basis, the Kriging interpolation method is used to complete the environment-dependent channel knowledge. Finally, the proposed CKM construction method is verified based on the simulations. Simulation results show that the sparse channel knowledge is well extracted and completed. The constructed CKMs are consistent with the reference CKMs.

There are task assignment problems in WSAN network. Coordination and real-time are two requirements that WSAN should meet, in which the main purpose of coordination between actuators and actuators is to accomplish the effective assignment of tasks. The actuator node is responsible for processing data and executing tasks, so when an event occurs, selecting the right actuator node can reduce energy consumption, speed up the response, and ensure real-time and energy balance, and extend the network life. In this paper, the cluster head node far from the Sink node consumes too much energy in single-hop communication, which reduces the network lifetime, and the “hot zone effect” caused by multi-hop clustering algorithm is discussed.

Hyperspectral unmixing technology is increasingly developing in the direction of artificial intelligence. Researchers are working on reducing the complexity of the unmixing network while improving the unmixing performance. In this paper, combined with the iterative formulation of the SUnSAL algorithm, an efficient unmixing network with interpretability is proposed. This method maps the SUnSAL algorithm to the convolutional neural network, and uses an attention mechanism for hyperspectral data, allowing the network to pay more attention to the calculation of important data. Specifically, we connect three variable update layers in series to realize iterative calculation, and the network scale is small and has strong interpretability. The experimental results show that the unmixing ability of the network is significantly better than the traditional sparse unmixing algorithm in three different of signal-to-noise ratios.

Hyperspectral sparse unmixing is an crucial preprocessing technique. In recent years, deep learning-based methods have gained increasing attention in spectral unmixing, with particular emphasis on the performance of unsupervised autoencoder (AE). To fully exploit both spectral and spatial information in hyperspectral bands for unmixing, this study explores a framework for multi-view spectral and spatial information unmixing based on autoencoder (AE). We incorporate multi-view spectral information by utilizing spectral partitioning and introduce a ConvLSTM encoder that leverages recurrent neural networks (RNNs) to synergistically exploit multi-view spectral and spatial information for more effective unmixing. The experimental results on synthetic datasets confirm the superiority of the proposed method in achieving excellent unmixing performance.

Due to the problem of low security in traditional car anti-theft systems, an intelligent remote anti-theft system for cars was designed with ARM chip TMS320DSC2X as the core control. Firstly, the overall structural and functional design of the intelligent remote anti-theft system was carried out. Then, the hardware structure and circuit design of the intelligent remote anti-theft system were carried out, including ARM chip circuit control system circuit design, wireless communication GSM module circuit design, voice module circuit design, automotive circuit control module design, and wireless remote control module circuit design. Finally, the software process design of the intelligent remote anti-theft system was carried out. The intelligent remote anti-theft system of this car can effectively handle detected abnormal situations in real time, driving the voice module to provide correct voice reminders, greatly improving the safety of car protection.

Multispectral images contain features with complex information and high spatial and inter-spectral redundancy. To better achieve multispectral image compression, an end-to-end multispectral image compression framework based on wavelet transform is proposed in this paper. In this method, the forward transform uses the boosted wavelet method and then the master encoder performs the mapping transform on the subbands after wavelet transform to obtain the latent features. The entropy model further extracts the potential features edge information for modeling to assist the entropy coding and decoding process of subbands, and finally recovered into multispectral images. The experimental results show that our method is feasible.

The In2O3/Au-nanorods (NRs) ethanol sensor with a simple peripheral driver circuit has been fabricated, which has highly sensitive to ethanol biomarkers in breath. This paper firstly describes fabrication method of In2O3/Au-NRs and its gas-sensing characteristics under laboratory conditions, then the STM32 sensor detection system and the application of the sensor in the system in detail. The In2O3/Au-NRs sensor exhibits strong humidity interference resistance and can detect different concentrations of ethanol gas accurately. The main controller STM32 is applied to the ethanol sensor data acquisition system to reduce the development cost and speed up the development cycles. A 4.3-inch capacitive touchscreen with integrated functionality modules including detection, alarm, touch input, and software reset has been designed, which has extensive versatility for ethanol detection scenarios. First, the principle and process of touch buttons, system block diagrams, hardware circuit schematics, and program flowcharts were designed. Then, the program development was streamlined and made more efficient with the help of STM32cubeMx and HAL libraries, simplifying the initialization and configuration procedures. The rendered images demonstrate that the ethanol detection system designed with In2O3/Au-NRs based on STM32 exhibits good detection accuracy and excellent portability.

Functional embedding in urban areas is the basis for human beings to move towards the construction of smart cities, and human aggregation and movement information can reflect urban functions. For any region, capturing human movement patterns and extracting the corresponding spatio-temporal features can effectively characterize the function of the city region and carry out city region function embedding. In our work, we consider the advantages of multi-graph attention network feature extraction and propose a new grid-based node-level multi-graph feature extraction method and capture road network features with N-level residual blocks, which in turn performs regional function embedding, called Multi-graph attention feature fusion segmentation network (MGAF-SegNet). MGAF-SegNet is a new multi-graph fusion residual network, in which the multi-graph attention mechanism can effectively capture the information within and between grids to mine the functional characteristics of urban areas. Considering the limitation of road network on urban functional division, incorporating the road network information, extracting the semantic relevance of the region, and realizing the division of urban functional regions from nodes to irregular regions through the semantic segmentation algorithm. The effectiveness of the proposed network is also demonstrated by taking Beijing as an example.

Extremely large-scale antenna array (ELAA) is a feature common to several key candidate technologies for sixth generation (6G) mobile networks. Due to the very large number of antennas in ELAA, the electromagnetic radiation field needs to be modeled with near-field spherical waves, which is different from the traditional plane wave-based radiation model of 5G massive MIMO. Therefore, near-field communications will become critical in 6G wireless networks, where the most critical is to delineate the boundary between the far-field and near-field regions, i.e., Rayleigh distance, also known as Fraunhofer distance. This paper summarizes the detailed derivation process for calculating the Rayleigh distance based on different criteria in existing near-field communications. Based on the phase difference criterion, the classical Rayleigh distance in MISO scenario and the MIMO Rayleigh distance (MIMO-RD) and MIMO advanced Rayleigh distance (MIMO-ARD) in MIMO scenario are introduced. The uniform-power distance (UPD) is introduced for the MISO scenario based on the amplitude/power ratio criterion of the signal between the array elements. Finally, based on the array gain criterion affecting the propagation rate, the effective Rayleigh distance (ERD) is introduced for the MISO scenario considering the angle factor. This paper provides a general understanding of the Rayleigh distance calculation for 6G near-field communications. The expected research directions and the work already done in the lab are summarized at the end of this paper.

Recently, intelligent reflecting surface (IRS) has been proposed and drawn significant research interests as a promising technique to enhance wireless communication performance. Since IRS technology is similar to the well-known relaying technology, it is necessary to compare the performance of the two different schemes. In this paper, we compare the IRS technology with classic decode-and-forward (DF) full-duplex relaying (FDR) from the perspective of spectrum efficiency (SE) and energy efficiency (EE). The analysis and simulation results show that IRS-assisted transmission outperforms DF-FDR-assisted transmission with a very large number of reflecting elements.

Locality Preserving Projection (LPP) is one of the most promising manifold learning methods, which is commonly used for feature extraction. But its purpose is only to preserve local distance information between samples, and does not consider the class information of samples that plays a crucial role in classification tasks, which leads to low recognition performance of LPP. In view of this, a new feature extraction method called Fuzzy Locality Preserving Projection (FLPP) is proposed for the classification of drone in this paper. The advantage of FLPP is that it can fully utilize the distribution information and the class information of the data samples by incorporating a fuzzy membership matrix into the neighborhood graph of LPP, and thus can improve the recognition performance. Experimental results show that the average recognition rate of the proposed FLPP is improved by 15.93%, 4.32%, and 3.25% compared with the Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA) and LPP methods, respectively.

Social network data contains a wealth of user behavior information, providing a basis for studying user preferences and information dissemination mechanisms in social networks. The high-dimensional and sparse nature of the data poses challenges for social network data analysis. In this paper, we focus on social network feature dimensionality reduction and analysis, and propose a comprehensive framework that integrates dimensionality reduction techniques for social network feature learning. The aim is to extract low-dimensional and efficient feature representations from complex social network data. This framework utilizes the neighboring relationships and similarity measures of nodes to construct features. It employs mainstream dimensionality reduction techniques to reduce the dimensionality of the data, thereby reducing the feature space while preserving critical information. Finally, a classification prediction model is built to accurately predict relationships between unknown nodes. Experimental results on multiple real social network datasets demonstrate that the algorithm proposed in this paper significantly improves the classification performance of social network data.

The home furnishing industry, closely linked to all, is undergoing tremendous changes due to the widespread consumption upgrades that have infiltrated all areas of the nation. Nowadays, more and more smart home products appear in the lives of ordinary people. Various smart products such as smart lighting, smart curtains, and smart sweeping robots have brought great convenience to people's lives. As an important part of smart home, smart door lock has also become an important choice for consumers to intelligentize their homes. Compared with the mechanical door locks on the market, smart locks are safer, smarter, and more humanized, and have been recognized by the market. In the smart door lock industry, many Fortune 500 companies are also making product layouts on smart door locks, such as China’s Xiaomi Company It produces smart door locks and is committed to building a smart home. The smart door locks in the market can support five unlocking methods, including fingerprints, mobile phones, access control doors, key passwords, and key unlocking functions. It can be said that it is an indispensable member of the smart home.

With the rapid development of urban traffic, the travel of the visually impaired is facing great challenges, and the traditional blind rod can no longer meet their needs for safe travel. At present, the development of front-end guide products is in a bottleneck period, and the application problems of traditional guide products are frequent, and there is a lack of innovation and ways to solve the problems. Learn from the popular intelligent devices in the last five years, thereby liberating the hands of the visually impaired, providing more accurate navigation services, improving the living standards of the blind, and ensuring the safety of the blind. In addition to visual impairment, blind people have the same intelligence as normal people. Normal travel is part of the independent life of disabled people, and they enjoy the same right to freedom as normal people. Providing travel freedom for the blind is conducive to equal participation in society and improving living standards. In order to solve the problem of difficult travel for the blind, we will design an intelligent guide cane, which is based on STM32 microcontroller and supported by GPS positioning module, ultrasonic ranging module, GSM module, voice module, etc., which can realize multiple functions such as real-time positioning, obstacle alarm, sending SMS, calling emergency contacts and so on. Walking stick adopts voice broadcasting and key operation to carry out man–machine interaction. The intelligent blind rod has complete functions, small size and low cost, and is easy to scale production.

In order to address the interference of concept drift on the results of multi-label learning algorithms, a hybrid kernel extreme learning machine is used as the foundation for the classification algorithm. Concept drift detection is incorporated, and the classifier is updated based on the detection results for application in multi-label learning. Firstly, the data stream is divided into appropriately sized data blocks, and a hybrid extreme learning machine is used on several of the preceding data blocks to obtain the base classifier. Subsequently, the incoming data blocks are processed using the base classifier to calculate the sample mean and variance between the current data and previous data. Based on this result, it is determined whether concept drift has occurred, and the base classifiers within the ensemble model are retrained and adjusted to update the model.

Multipath angle characteristics have a significant impact on the performance of the air-to-ground (A2G) wireless communications. This paper proposes a modified clustering algorithm based on a multipath cluster model to analyze extracted parameters, such as delay and angle, from realistic A2G channels in urban scenarios. A detailed scheme is presented for investigating the multipath angle characteristics in different scenarios at varying heights. This proposed scheme includes the reconstructed digital map of a realistic urban scene and A2G communication simulation by using the ray tracing (RT) method. Additionally, the study introduces a Gaussian mixture model (GMM) that fits the cluster angle offset well and offers broad applicability. The simulation results demonstrate the effectiveness of the proposed clustering algorithm in analyzing A2G channel multipath parameters.

Unmanned aerial vehicle (UAV) is becoming an effective solution for collecting IoT data. However, due to its limited battery capacity, UAV cannot complete data collection tasks over broad areas or a long time, which is incompatible with attaining fairness and high energy efficiency in data collection. To address the above challenges, the intelligent reflecting surface (IRS) is introduced as a solution. It can enhance communication by separately controlling the phase shift of each element. This paper investigates the problem of the IRS assisting a recharged UAV for data collection. We propose a proximal policy optimization (PPO)-based algorithm to jointly optimize the phase shift of IRS and the flight trajectory of UAV. To prevent crashes, we allow the UAV to return to the charging station when its battery is lower than the threshold. Simulation results show that the proposed method outperforms existing solutions in terms of fairness and energy efficiency.

To improve the accuracy of plant leaf disease image recognition, a CVT-based image classification algorithm is proposed. The algorithm utilizes Convolutional and Transformer networks for feature extraction and encoding, integrating global and local image features. By introducing the self-attention mechanism of Transformer, the algorithm achieves weather image data classification. Experimental results demonstrate that the CVT-based deep learning algorithm effectively enhances model prediction accuracy, showing promising results in plant leaf disease recognition. The algorithm achieves accurate recognition of five different classes of data, with an accuracy rate as high as 97.78%.

The present paper deals with an overview of particle swarm algorithms and hybrid genetic-particle swarm algorithms, and a comparison between the related convergence speed and correlation error. The work includes also a dynamic mutic-objective charging model that is important for the security, stability, and economics of the smart grids. Finally, a brief analysis of GA-PSO multi-objective electric vehicle charging dynamic optimization is reported.

This paper proposes a pricing strategy for the Market by using a Stackelberg game-based bi-level programming model. In the model, the wholesale price and the retail price is optimized by the Market to increase its profit. In the model, the Energy Provider (EP) optimize optimizes the adjustment coefficient to increase the trading probability with the Utility Company (UC). Besides, the Market can sell the power to the UC directly. Then, the UCs determine their retail price by non-cooperative game to increase their profit. The simulation results reveal that the Market determines the optimal price to maximize profits. On the other hand, the proposed price strategy of EP can increase the trading probability, which promotes more UCs to trade with EP, thus increasing its benefit. Finally, UCs get a bash equilibrium through the non-cooperative game. The research proves that the proposed strategy is a win–win strategy.

As a high-demand product with increasing frequency of use, keyboard has a broad space for exploration and innovation. In order to make human-computer interaction more convenient, faster and more efficient, a new generation of input devices with keyboards as the carrier was developed here. The new generation of input devices not only provides more efficient and accurate input, but also more functions and ways to interact. The new generation of input devices makes the input more intelligent and humanized by combining different technologies, and provides more choices and conveniences to meet people's input needs. At the same time, there are still many places where innovation can be made, such as speech recognition input. This input method can realize functions such as voice assistant, voice search, voice translation, etc., and has high convenience. Smart keyboards with such functions are bound to bring a more efficient and comfortable input experience, and in order to improve keyboard life and input accuracy, it is also necessary to design corresponding key stabilization circuits and electrostatic protection circuits.

In the past few decades, with the continuous progress in devices and research, both hyperspectral imaging technology and three-dimensional reconstruction techniques have made significant advancements. In recent years, three-dimensional reconstruction based on hyperspectral imaging has shown remarkable results. Unlike traditional three-dimensional reconstruction techniques based on RGB images that only capture the geometric information of the target object, the spectral diversity provided by hyperspectral images allows for the reconstruction of three-dimensional models with higher accuracy and better visual effects. Consequently, an increasing number of researchers have started to explore the field of three-dimensional reconstruction based on hyperspectral imaging. This article provides a comprehensive review of the specific steps involved in three-dimensional reconstruction based on hyperspectral imaging, including traditional algorithms used in the reconstruction process. Furthermore, it introduces the latest research on implementing three-dimensional reconstruction of hyperspectral images with the assistance of deep learning techniques.

Rotary steerable drilling rigs are common equipment in the field of oil exploration and production. Precisely controlling the posture and rotation direction of the drill bit is crucial for improving drilling efficiency and ensuring drilling safety. Deep learning technology, known for its powerful learning capabilities, has been widely applied in various fields. Applying deep learning techniques to rotary steerable drilling rigs holds significant importance in enhancing drilling quality and efficiency. In this paper, we propose a deep learning-based method for calculating angle errors. By using the improved differential contour method to calculate the rotational angle error of the rotary steerable drill bit, automatic control of borehole direction and depth can be achieved. This approach improves drilling quality and efficiency while reducing drilling safety risks.

Private-Internet of Things (P-IoT) is IoT network with for private users using dedicated frequencies. Ad hoc mode is the main operation mode of P-IoT. Since Ad hoc mode of P-IoT is a narrowband network, the congestion is the fatal problem on the intermediate nodes. Topology control is an effective way to reduce congestion. A topology control algorithm based on node degree for P-IoT is proposed in this paper. The topology control algorithm minimizes the node degree of the communicating nodes under the premise of guaranteeing network connectivity and constructs a network topology with less sparsity. The simulation results show that our algorithm can markedly reduce the average node degree and the sparsity of topology without decline of the network connectivity.

With the rapid development of deep learning technology, its applications in various fields are also increasing. In addition to making gratifying progress in traditional image classification, speech recognition, text classification, and other fields, it has also begun to play an important role in more specific and professional research scenarios, such as applying it to specific work such as fault detection in the power industry and case text analysis in the public security field, Fully utilize its self-learning and self-improvement characteristics and functions to provide assistance for the implementation of practical work. Based on the characteristics of deep learning technology, this article starts with the study of classroom teaching behavior analysis in the field of teaching analysis, and explores the feasibility of applying deep learning technology to classroom teaching behavior analysis research.

With the continuous advancement of the construction of smart cities, the efficient acquisition and automatic extraction of building information is very important. Building extraction based on high-resolution remote sensing images is an important subject in current remote sensing technology. This paper summarizes the building extraction methods of high-resolution remote sensing images, describes them from traditional methods and deep learning-based methods respectively, and summarizes the evaluation indicators, advantages and disadvantages and application scope of each method. The potential of automation, efficiency and precision of high-resolution building extraction in the future is also discussed.

With the rapid development of drone technology, drone imagery has become an important means of obtaining high-resolution surface information. However, due to the operational height and range limitations of drones, there are issues of limited coverage and small data volume in drone imagery. Meanwhile, satellite imagery offers extensive coverage and a large amount of data but with lower resolution. In order to fully utilize the advantages of drone imagery and satellite imagery, and improve the accuracy of surface information extraction and spatial resolution, researchers have conducted studies on the fusion algorithms of drone imagery and satellite imagery. This article provides a review and analysis of the fusion algorithms for drone imagery and satellite imagery. Firstly, the characteristics and advantages of drone imagery and satellite imagery are introduced, emphasizing the importance of integrating the two. Furthermore, data loading and preprocessing techniques are discussed. Then, common fusion methods for drone imagery and satellite imagery are detailed, including pixel-level fusion, feature-level fusion, and decision-level fusion, among others. The evaluation methods for fusion quality are also explained. Finally, research achievements from both domestic and international sources are presented.

In this paper, a temperature control system based on micro controller unit is developed to obtain the operating temperature of the semiconductor laser by measuring and calculating the negative temperature coefficient thermistor through the microcontroller STM32F103ZET6, and then a direct proportional-integral-derivative (PID) algorithm is used to accurately control the preset operating temperature. According to the experimental verification obtained: this semiconductor laser temperature control system can accurately and efficiently control the working temperature at about 25 °C, floating range of ± 0.1 °C. Compared to similar products, this system demonstrates higher accuracy and extremely fast response rate in wide temperature environments.

With the development of “Digital Earth,” “Reality-Based 3D China,” and “Smart Cities,” technologies such as UAV aerial photography, photogrammetry, LiDAR, oblique photogrammetry, and SLAM are increasingly utilized for constructing reality-based 3D models. However, each individual technology has its limitations in 3D reconstruction, especially in the fine modeling of buildings, such as occlusion in texture mapping and low model accuracy. While oblique photogrammetry captures multi-angle images of terrains, it encounters challenges when acquiring images in concealed locations, resulting in structural deformations and surface artifacts, leading to insufficient model precision and poor elevation accuracy. LiDAR point clouds complement the geometric structure in the blind areas of oblique photogrammetry, resulting in smoother ground surfaces and sharper edges and lines at the base of buildings. Integrating vehicle-mounted LiDAR point clouds with oblique photogrammetry effectively compensates for the limitations of using a single data source for 3D model creation and improves model accuracy. In 3D reconstruction, the fusion of oblique photogrammetry and LiDAR data is crucial. The classical Iterative Closest Point (ICP) algorithm, widely used in point cloud registration, iteratively finds the closest point pairs between two point sets to calculate the transformation matrix, converging to a certain threshold. However, ICP requires a high initial position accuracy of point clouds, and its simple selection of corresponding points based on Euclidean distance may lead to mismatches, impacting the registration precision. As a result, numerous scholars have made improvements and research on this algorithm.

In order to provide a more scientific rehabilitation training plan for patients with lower limb paralysis caused by illness or accidents when they train independently or at home, a lower limb rehabilitation training system with myoelectric assessment was hereby designed to improve the therapeutic effect and to reduce the difficulty of nursing care. The collected electromyographic signals of the target muscle surface after homogenization and standardization were used as the evaluation standard, and suitable training plans were given for different muscle functional states of different patients, which were combined with the mobile terminal to realize the visualization of data and training guidance. According to the repeated experiments, the accuracy of the judgment of the functional status of the target muscle of the patient could reach more than 85% without resistance, and the judgment accuracy was increased to more than 92% after adding resistance appropriately. The design can directly feedback and guide rehabilitation training through EMG and visual interface, which will further improve the effectiveness and convenience of training for patients with lower limb dysfunction.

The conductivity is determined by the number of conductive particles attached to Thin Film Transistor Liquid Crystal Display (TFT-LCD) bumps. The image detection method for the number of particles is widely used to judge the quality of TFT-LCD bumps. We propose a detection method of CNN-based Classification for conductive particles in TFT-LCD manufacturing. First, aiming at the phenomenon that the adhesion of conductive particles seriously affects the detection accuracy, a binary classification detection of bumps conductivity is proposed based on a static number of particles attached to the bumps. Second, a convolutional neural network (CNN) classification model is established; due to the bump images with multiple gray levels, principal component analysis (PCA) feature selection is introduced into the classification model. This model can learn features of particle from massively labeled data. The experimental results show that the accuracy of our proposed method is 95.55%. The accuracy is 1.46% higher than the K-means clustering + CNN method, 2.45% higher than the watershed method, 3.00% higher than the K-means clustering method and 7.70% higher than the Otsu method. The method proposed in this paper can be effectively applied to the quality detection of TFT-LCD bump.

Convolutional neural networks using deep learning can automatically identify and locate disease markers, improving the accuracy of diagnosis by doctors. In deep learning models, YOLO, as a representative of single-stage models, has the advantages of high accuracy and high detection speed, and has been widely used in the field of object detection. YOLOv5 uses deep learning technologies such as anchor boxes, focal loss, and data augmentation, combined with improvements such as Dark and SPP algorithms, and has the characteristics of a small model and fast speed, making it suitable for deployment on mobile devices. Based on the YOLOv5 model, this article builds a lightweight model for detecting fracture locations in X-ray imaging. This model can assist doctors in diagnosis, reduce misdiagnosis rates, and has important application value.

With the shrinking of the feasible domain, the problems of slow convergence and easy to fall into local optimum in Particle Swarm Optimization (PSO) algorithm become more serious. To solve this problem, the paper presents an improved PSO method. First, The mass of the initial population is optimized by using an opposition based learning strategy. Secondly, The algorithm is divided into two stages according to the feasible solution proportion of the population. The improved algorithm is applied to synthesize directional modulation signals on a time-modulated array (TMA). Experimental results show that the proposed algorithm has better convergence speed and optimization ability.

UAV with a beamforming design as a flight base station to assist communication is a promising solution to alleviate the burden and compensate for the shortcomings of the communication infrastructure. However, when UAVs are deployed as flying base stations in a fixed range to provide signal transmission services for users, their performance analysis and optimization problems have not been well studied. In this paper, the system model was established, and the optimization scheme was proposed. Then, based on the mathematical analysis results, the system performance was further improved by adjusting the altitude of the UAV, and the UAV deployment problem was formulated as the problem of maximizing the signal power received by the user. To find the optimal solution, a method combining exhaustive search and an artificial bee colony (ABC) algorithm was proposed to solve the corresponding optimization problem of UAV altitude under beamforming. The effectiveness of the theoretical analysis and the proposed scheme was verified by numerical simulations.

A directional modulation (DM) based dual function radar and communication system is developed in this paper. By solving the non-convex optimization iteration problem, four weight vectors are designed, and each weight vector controls a waveform. QPSK phase modulation is realized in the communication direction and phase value is disturbed in other directions without affecting the main radar functions. And the effectiveness of the design has been proven through simulation results.