Abstract:It has become a mandatory requirement for electric bike riders to wears helmet on riding.To automatically check if the electric bike rider wears helmet,a helmet and license plate detection approach based on improved YOLOv5m model is herein proposed,which can locate and recognize the license plate of the unhelmeted rider,so as to track down the violators.The model is trained with self-built dataset,uses DIOU loss function instead of GIOU loss function,and uses DIOU_NMS to replace weighted NMS so as to enhance the recognition ability for dense cycling scenes.Meanwhile,the ECA attention mechanism is added to the Backone and the Neck parts to improve the recognition accuracy for small- and medium-sized targets.Then,the K-means algorithm is used to re-cluster the anchor frame size.Finally,the Mosaic data enhancement method is improved.The experimental results show that the mAP of the proposed approach is 92.7％,which is 2.15,5.7,and 6.9 percentage points higher than the original YOLOv5m,YOLOv4 tiny,and Faster RCNN,respectively.It can be concluded that the improved YOLOv5m model can accurately recognize rider's helmet and electric bike's license plate.

Abstract:An algorithm based on improved YOLOv5s is proposed to address the problems of small percentage of traffic signs in the image,low detection accuracy and complex surrounding environment.First,the attention mechanism of ECA (Efficient Channel Attention) is added to the backbone network part to enhance the feature extraction ability of the network and effectively solve the problem of complex surrounding environment.Second,the HASPP (Hybrid Atrous Spatial Pyramid Pooling) is proposed,which enhances the network's ability to combine context.Finally,the neck structure in the network is modified to allow efficient fusion of high level features with underlying features while avoiding information loss across convolutional layers.Experimental results show that the improved algorithm achieves an average detection accuracy of 94.4％,a recall rate of 74.1％ and an accuracy rate of 94.0％ on the traffic signage dataset,which were 3.7,2.8,and 3.4 percentage points higher than the original algorithm,respectively.

Abstract:To address the problem that the railway safety zone division along the electrified railway with complex background needs to use actual fixed standard parts as reference and the division range is small,a smart safety zone division approach independent of reference objects is proposed.The GSD (Ground Sample Distance) parameters are calculated from relevant parameters in images collected by UAVs (Unmanned Aerial Vehicles),and the DeepLabv3+ model with ECA-Net module is used to accurately segment the railway in the image.Then,a series of image processing operations such as edge detection,opening operation,and probability Hough transform are used to extract the key pixel points that make up the railway,and the least squares algorithm is used to fit the railway and obtain its mathematical expression.Finally,mathematical models,GSD parameters,and the mathematical expression of the railway are combined to complete the safety zone division.Experimental results show that the proposed approach achieves measurement accuracy over 90％,doesn't need to select fixed reference objects,and has strong adaptability and high robustness.The high practicality and reliability of the proposed approach provides effective technical support for safety management along the electrified railway.

Abstract:The rapid development of network technology,human-computer interaction and artificial intelligence has given birth to the metaverse and further promoted the digital transformation of all aspects of people's life.The concept of metaverse emerged in 2021,and has attracted extensive attention from industry,academia,media and the public.This paper attempts to deeply analyze the metaverse from the perspective of technology and application.First,the concept and connotation of the metaverse are elaborated from the definition,origin & development,characteristics and key technologies (network and computing,Internet of Things,human-computer interaction,electronic game,block chain,digital twin,etc.).Then,the enterprises and application examples involved in the metaverse are discussed.Finally,the existing challenges and opportunities in the development of the metaverse are analyzed,and the future researches and applications are prospected.Through the meta-analysis of the current development status and research trend of the metaverse and the scientific evaluation of the potential application of the metaverse,we can provide useful reference for the researchers of the metaverse.

Abstract:To address the problem of numerous hyperparameters,loss of long time series information and difficulty in distinguishing primary and secondary features in Long Short-Term Memory network (LSTM) for grain yield capacity prediction,this paper proposes a combined data-driven grain capacity forecasting model.In the hyperparameter part,the proposed model performs hyperparameter search optimization for LSTM by introducing Dynamic Weights and Laplacian variation of Bald Eagle Search Optimization Algorithm (WLBES),to avoid the process of manual parameter adjustment.In the prediction part,the proposed model uses Ridge Regression (RR) to correct the residuals of the prediction results to make up for the deficiency of LSTM data loss,and adds an attention mechanism to distinguish primary and secondary features by weight size to enhance the importance of features with greater relevance to grain production.The results show that the combined WLBES-LSTM-RR model decreases the root mean square error (RMSE) by 75％ and 19％ compared with the LSTM and WLBES-LSTM models,respectively,and substantially decreases the RMSE compared with other combined models of optimized LSTM.This combined model has higher prediction accuracy in grain yield capacity prediction.

Abstract:Air pollution seriously endangers the travel safety and health of residents.As a comprehensive indicator used to measure air quality condition,Air Quality Index (AQI) can alert the public to air quality and enable people to make more informed travel decisions.By predicting the change of air quality in advance,the government and environmental protection departments can take emergency measures to reduce air pollution.Here,we propose an integrated deep learning model based on Convolutional Neural Network and Gated Recurrent Unit (CNN-GRU) for AQI prediction.The CNN is used to extract the spatial and temporal characteristics of air pollutants and AQI and complete the feature mapping,while the GRU to model the temporal relationship and complete the calculation and AQI efficiently.The daily average concentrations of six major air pollutants (PM_2.5,PM₁₀,SO₂,CO,NO₂,O₃) in Beijing and Guangzhou during 2014-2022 are selected for example study,and the AQI is predicted using the CNN-GRU model.The results show that,compared with Multiverse-Optimized Generalized Regression Neural Network model (MVO-GRNN) and Genetic Algorithm-optimized BP neural network model (GA-BP),the proposed CNN-GRU model has the smallest prediction error for AQI.

Abstract:Indices such as tumor cell density,nucleocytoplasmic ratio,and average size have important implications for cancer grading and prognosis.Therefore,segmentation of nuclei is the fundamental prerequisite for tumor microenvironment analysis in computational pathology.Additionally,the exploration of new tumor markers is of great significance through statistical analysis of segmentation results.However,the morphology of nuclei in the background of pathological images is irregular,the staining of nuclei is uneven,and adhesion occurs between the edges of nuclei.While the segmentation error of the edge will make no difference on the overall loss as long as the main body of the nucl is correctly segmented,so the adhering nuclei can easily be regarded as the same segmentation target by existing deep learning algorithms.To address the overlapping nuclei,a new segmentation algorithm based on the Transformer and distance map,abbreviated as TDM-Net,is proposed,which integrates the core of multi-head self-attention mechanism in Transformer with contextual information to fully explore the proximity relationship and enhances the learning ability of image details by introducing distance map to emphasize the interior of nuclei and weaken the boundary of nuclei.The algorithm's Dice coefficient,precision,Aggregated Jaccard Index (AJI) and Hausdorff distance are 0.797 9,0.756 1,0.667 2,and 10.11,respectively.The results show that the proposed TDM-Net outperforms other segmentation algorithms,effectively improves nuclei segmentation accuracy and solves overlapping of different nuclei.

Abstract:Aiming at the problems of boring contents of traditional wrist rehabilitation training methods and low training efficiency due to users' lack of motivation to participate,a wrist training system of myoelectric control virtual reality game was designed.The surface electromyography (sEMG) signals of wrist movement were collected and the wrist joint movement intention was decoded through the principle of muscle synergy for the control of the virtual reality game;random disturbance force was introduced into the virtual reality game,and the interaction with the virtual reality environment was realized through the way of impedance control,which enabled users to explore different movement control methods.The feasibility of the system was verified through model calibration experiments,and training experiments were conducted to assess the training effect by evaluating the task completion time as well as the path efficiency.The experimental results show that introducing random interference force reduces the task completion time by 24％ and improves the path efficiency by 26％,and the designed training system enables users to perform motion control in a more efficient way and improves the training efficiency.

Abstract:Pervasive edge computing allows peer devices to establish independent communication connections,which enables users to process massive computing tasks with low delay.However,distributed devices cannot obtain the global system status of the network in real time,thus the fairness of resource utilization cannot be guaranteed.To solve this problem,a resource allocation scheme for pervasive edge computing based on Generative Adversarial Network (GAN) is proposed.In this scheme,a multi-objective optimization problem is established for minimizing the time delay and energy consumption,which is then transformed into a maximum reward problem according to the random game theory.And then a computation offloading algorithm based on multi-agent imitation learning is proposed,which combines multi-agent Generative Adversarial Imitation Learning (GAIL) and Markov Decision Process (MDP) to approximate the performance of experts,and realizes online execution of the algorithm.Finally,combined with Non-dominated Sorting Genetic Algorithm Ⅱ (NSGA-Ⅱ),the time delay and energy consumption are jointly optimized.Simulation results show that,compared with other edge computing resource allocation schemes,the proposed solution shortened the time delay by 30.8% and reduced the energy consumption by 34.3%.

Abstract:Sonar images are prone to problems such as low contrast,low resolution,and edge distortion,so it is difficult to accurately separate effective signals from noise when removing noise from sonar images,resulting in reduced image contrast,unclear edge contours,and severe detail loss after denoising.Therefore,this paper proposes a sonar image denoising algorithm based on adaptive Wiener filtering and 2D-VMD (Two Dimensional Variational Mode Decomposition).First,a noisy image is decomposed using 2D-VMD to obtain a series of sub modes with different center frequencies.Effective modal components are obtained via correlation coefficients and structural similarity,then processed by adaptive Wiener filtering,and finally the filtered modal components are reconstructed to remove noise.The experimental results show that the proposed image denoising algorithm achieves the best results in terms of correlation coefficient and structural similarity,with a peak signal-to-noise ratio slightly lower than that of NSST domain denoising.Taking into account objective data and visual effects,the algorithm proposed in this paper achieves the best performance in image details and edge preservation after removing noise.

Abstract:To address the problems of high cost of time and labor and low efficiency frustrated traditional remote sensing image processing,an improved 2DCNN (2D Convolutional Neural Network) model abbreviated as En-De-2CP-2DCNN was proposed,with the purpose to improve the processing speed,accuracy and reduce the number of parameters in the classification of remote sensing Hyperspectral Images (HSI).First,1DCNN,2DCNN and 3DCNN were used to carry out classification experiments on Pavia University HSI dataset,and their advantages and disadvantages were compared and analyzed.Second,under the premise of maintaining fast processing speed without increasing model parameters,the 2DCNN was selected as the basic model,which was then improved with referring to the Encoder-Decoder structure of SegNet and integrating the idea of double convolutional pooling,and the learning strategy was optimized.The results show that the F1-score of the proposed En-De-2CP-2DCNN model is 99.96％,reaching the same level of 3DCNN (99.36％),which is 2.68 percentage points higher than that before improvement (97.28％);the processing speed (5 s／epoch) is comparable to that of 1DCNN and faster than 3DCNN (96 s／epoch);the amount of parameters is reduced from 3.55 MB to 2.01 MB,which is higher than 3DCNN (316 KB) but much lower than 1DCNN (19.21 MB).The proposed En-De-2CP-2DCNN model realizes accurate,fast and lightweight processing of remote sensing hyperspectral images.In particular,the improvement in processing speed and parameter amount is conducive to further realizing the lightweight deployment of mobile terminals.

Abstract:Global Navigation Satellite System (GNSS),Strapdown Inertial Navigation System (SINS) and visual sensors can complement each other,and their information fusion can obtain high-precision,drift-free navigation and positioning information.Aiming at the problem that GNSS／SINS／vision fusion navigation is vulnerable to the impact of motion speed,light change,occlusion,etc.,which leads to the decline of navigation positioning accuracy and robustness,this paper adds the SoftLone robust kernel function to the cost function of the graph optimization framework,and sets the gross error test procedure of the measured value to reduce the negative impact of outliers.Further,the chi-square test is performed on the calculated residuals of the measured value,and the weight of the over-limit residual is reduced to improve the accuracy and robustness of the system.The experimental results show that the proposed algorithm has higher accuracy and better robustness than traditional algorithm without robust kernel function,outlier elimination strategy and chi-square test,and algorithm with other robust kernel functions.It can greatly improve the positioning accuracy and robustness of GNSS／SINS／visual navigation.In large scale scenario,there is no large drift errors,and root mean square error and standard deviation of absolute pose are 0.735 m and 0.336 m,respectively.

Abstract:The track geometry state measuring instrument using GNSS positioning cannot work in GNSS rejection environment such as tunnel and underground.Since the railroad track is still close to the design alignment even if it is deformed,the difference between actual track position and its design value always remains within a certain range.Herein,a track irregularity detection approach considering design attitude assisted IMU／ODO (Inertial Measurement Unit／Odometer) is proposed,which combines design attitude with inertial guidance solved attitude for Kalman filtering and uses odometer velocity for dead reckoning.Computational experiment results show that the railroad design attitude information can significantly improve the track irregularity detection accuracy,and the proposed approach is comparable to the total station assisted dynamic detection method according to the 30 m chord track irregularity detection results.The overall detection efficiency is high enough to meet the needs of daily track detection.

Abstract:To explore the illumination adaptability of environmental perception equipment in application of SLAM (Simultaneous Localization And Mapping),comparative experiments of lidar and depth camera for SLAM were carried out under different illumination intensities.Combined with the LOAM (Lidar Odometry And Mapping) and RTAB-MAP (Real-Time Appearance-Based Mapping) algorithms,a 16-line lidar and a depth camera were placed on a four-wheel differential robot to carry out SLAM application in bright and dark environments.The experimental results show that in bright environment,the median errors of system deviations are 0.203 m and 0.644 m for the visual SLAM and lidar SLAM,respectively,which are 0.282 m and 0.683 m respectively in dark environment.The depth camera outperforms the lidar in positioning and mapping performance in both bright and dark environment,and it can be concluded that the depth camera is more illumination adaptable.

Abstract:With the completion of the BDS-3,China has become the third country that has global navigation satellite system.Furthermore,BDS-3 can broadcast observation signals up to five frequencies,which has great significance to achieve rapid fixing of ambiguity and improve positioning accuracy.The Ionosphere Reduced (IR) combination has minimal ionospheric delay and integer characteristics,therefore,a solution model based on IR combination is established for medium and long baseline utilizing the quad-frequency observations of BDS-3 in this paper.In the meanwhile,the tropospheric delay is estimated with tropospheric mapping function.The experimental results show that compared with traditional dual-frequency Ionosphere-Free (IF) model,the ambiguity fixing speed of the IR model is improved by more than 10％,and the positioning accuracy in north (N),east (E) and up (U) directions are improved by 7.7％,7.9％ and 8.2％,respectively.