Abstract:Brain-computer interface (BCI) technology aims to establish a new communication and control channel between the brain and the external environment that does not depend on peripheral nerves and muscles. The steadystate visual evoked potential (SSVEP) based braincomputer interface (BCI) is currently the noninvasive BCI paradigm with the highest information transmission rate, but it is still lower than the traditional interaction mode. In this paper, a hybrid braincomputer interface (BCI) combining surface electromyography (sEMG) and steadystate visual evoked potentials is proposed to further improve the information transmission rate of the system. A hybrid BCI system was realized by combining the SSVEP encoding at different frequencies with sEMG. The canonical correlation analysis method is used to identify the frequency of SSVEP signal, and the frequency domain analysis method is used to detect sEMG signal. Offline results from 8 healthy subjects show that the system can achieve an average accuracy of 84.28% and an average information transfer rate of 72.63 bits/min. These results lay the foundation for hybrid braincomputer interface studies combining surface EMG and steadystate visual evoked potentials.

Abstract:The aim of this study is to address issues such as the large volume of realtime measurement data for pipeline vibrations, prolonged transmission delays, and wastage of computational resources. By adopting edge computing theory, the data processing steps are moved closer to the devices, thereby accelerating the speed of status monitoring and optimizing the utilization of computational resources. This study provides a detailed overview of the overall functional framework, hardware design methodology, and vibration signal conversion algorithm of the edge computing perception system. The system consists of two parts: edge computing devices and data aggregation devices. The former is positioned at mechanical vibration sources for realtime analysis and processing of extensive redundant data, while the latter communicate with multiple edge computing devices via wireless signals to project aggregated information to maintenance terminals. Using a phaseshifting camera pipeline vibration experiment as a case study, this research demonstrates that the structural vibration perception technology based on edge computing can accurately identify abnormal vibration phenomena in the pipeline at frequencies of 64 Hz, 115 Hz, and 279 Hz. By guiding the use of vibration dampers, the study achieved a significant reduction in the pipeline′s maximum vibration amplitude from 068 m·s-1 to 00016 m·s-1, showcasing its substantial engineering practical value.

Abstract:Low PUE, low WUE and low CUE have become key approval indicators and economic indicators for data center construction. Many scientific research institutions and enterprises around the world are exploring relevant implementation technologies and solutions. In this paper, based on wind plant offshore substation a new data center implementation technologies with PUE<11, WUE≈0 and CUE≈0 is provided, by optimizing offshore wind power, marine circulating water and liquid cooling server through AI technology. According to the analysis of the pilot project, the PUE of this technology can reach 1067. Through joint design, the new energy and new infrastructure are fully reused, with outstanding economic benefits. It can be replicated and promoted on a large scale in the southeast coast of China, and can solve the problem of providing nearby computing power in the eastern region under the dualcarbon goal.

Abstract:The accurate measurement of the ground capacitance is helpful to ensure the compensation effect of the arc suppression coil and reduce the damage of the fault to the distribution network. A unified measurement method for ground capacitance and arc suppression coil inductance of resonant grounding system based on signal injection method and fast Fourier transform is proposed. Firstly, the relationship between the neutral voltage and the system parameters is established. Secondly, the nonpower frequency current source is connected to the neutral point, and the current source frequency component of the neutral point is obtained by using the fast Fourier transform to eliminate the power frequency error. Then the unified measurement of ground capacitance and arc suppression coil inductance is realized according to the equivalent circuit. Finally, the simulation model is built by Simulink, which shows that this method is suitable for various types of arc suppression coils. In the threephase asymmetry condition, the maximum errors of ground capacitance and arc suppression coil inductance are 0.11% and 0.1% respectively, and the maximum errors are 1.3% and 1.8% respectively in threephase symmetry condition.

Abstract:Liquid pharmaceuticals filled with vials are widely used and stored in hospitals and pharmacies. In order to achieve realtime monitoring of such agents, an on-line monitor was designed with photoelectric sensor, STM32 main control chip and UDP communication protocol. The light intensity parameters of the photoelectric sensor used in the monitor are calculated according to the inverse square law of direct illumination distance of point light source, and verified by Comsol simulation platform. The monitor uses photoelectric sensors to monitor the number of drugs in real time and synchronize drug information. The designed PCside drug monitoring software uses UDP protocol to connect multiple drug monitors to achieve onetomany online monitoring functions. The monitor can use a variety of power supply methods to monitor the ambient temperature in real time, and meet the needs of various types of vials filling liquid pharmaceuticals in multiple scenarios and multiple environments. The test results show that the power consumption of the monitor is 3.08 W, 3.23 W and 3.35 W at no-load, half-load and full-load, respectively, and the monitoring accuracy of the remaining number of vials is 100%. Combined with the low cost features, the monitor is worth promoting in hospital pharmacies, retail pharmacies and other places.

Abstract:This paper studies the radar servo multi-channel communication platform of domestic devices. Aiming at the problems of autonomous control of radar servo system hardware and complex multichannel communication, a domestic radar servo multichannel communication system based on MCU+FPGA architecture is proposed to improve the reliability, redundancy and universality of servo control. A 6way RS422 fusion structure is designed to complete the communication with each subsystem component, In addition, the device schematic design and PCB layout and wiring are completed on Altium Designer, and the 6layer stack design and physical static test of PCB are realized. The test results show that the communication system of this design meets the design requirements and can realize domestic substitution.

Abstract:Due to the electromagnetic interference, the onboard electronic measuring equipment will cause malfunction or malfunction. The traditional chaotic spread spectrum can reduce the electromagnetic interference of the onboard LLC resonant converter from the source, but with the increase of the spread spectrum width, the output voltage ripple will increase and the efficiency will decrease. Based on this, firstly, the chaotic sequence is generated by a lattice multivortex chua′s system, and the chaotic spread spectrum is realized in the way of leftward spread spectrum. Secondly, proposing a segmental traversal optimization algorithm with the rate of change of electromagnetic interference rejection and the rate of change of spreading width as the criterion, combine the prediction model of conducted interference with highfrequency parasitic parameters, and perform the spreading width optimization for different frequency bands specified in the national standard. Finally, the simulation and experimental results verify that the proposed strategy can achieve the maximum 7.07 dB EMI reduction, the maximum 61.5% output voltage ripple reduction, and the circuit efficiency improvement.

Abstract:Channel materials and environment in communication system will affect the phase of RF signal. Aiming at the problems of low calibration frequency and large error in traditional phase calibration methods, a phase difference calibration method based on software radio technology is proposed. The phase value of the maximum spectral line is solved by the fast Fourier transform and the coordinate rotation digital computer, and the phase calibration of the signal is completed by combining the digital orthogonal technique. The method was tested experimentally with the billiongate FPGA as the core. Xilinx Artix7 was used to generate reference and to be calibrated signals, and logic circuits such as DDR3 storage, phase difference measurement and calibration of billiongate FPGA were designed to complete phase difference calibration. The test results show that the relative error of phase difference is less than 03% and the rootmeansquare error is only 007° for radio frequency signals with a frequency of 1 GHz in the range of -90° to 90°.

Abstract:The coiling temperature control accuracy and coiling hit rate of hot strip rolling are low due to the influence of strong nonlinearity and time variation. This paper proposes a method to optimize long shortterm memory (LSTM) neural network based on improved whale algorithm. The improved Whale optimization algorithm of niche technologymixed mutation strategy (NMWOA) was obtained by combining adaptive parameter optimization and hybrid mutation strategy with niche technology. The coiling temperature prediction model of LSTM optimized by improved whale algorithm was established and compared with other models. Simulation results show that the NMWOA algorithm has better search ability and optimization accuracy among the 10 test functions compared with other advanced algorithms. In the prediction of the coiling temperature model, compared with the other four models, the NMWOALSTM model has a high precision hit rate of 9750%, which improves the prediction accuracy of the coiling temperature.

Abstract:In order to improve the impact of vertical, pitch and roll motion of the vehicle in the process of driving, a multiobjective switching control strategy of active suspension is proposed. Based on the extended zero moment point theory, the pitch and roll evaluation indexes of the vehicle in complex terrain are established. The vehicle model is established in ADAMS/Car environment, and the accuracy of the model is verified by real vehicle tests based on random roads. According to the graphical evaluation index and the steering wheel Angle value as the logical judgment condition of state switching, a multiobjective switching control strategy is established on MATLAB/Stateflow, and a multiobjective switching fuzzy PID controller is established based on the control strategy. A certain type of linear motor is selected as the force source of active suspension, and the cosimulation under three different conditions of constant speed driving on Cclass road, double line shifting, acceleration and deceleration up and down slope is carried out by ADAMS/Car and MATLAB/Simulink respectively. The simulation results show that the proposed multi-objective switching control strategy of active suspension effectively reduces the occurrence of vertical, pitch and roll motion of the vehicle, and improves the ride comfort and driving safety of the vehicle as a whole.

Abstract:Under the condition of non-uniform temperature field in thermoelectric power generation system, the power voltage curve has multipeak characteristics, the traditional particle swarm optimization algorithm is easy to fall into the local optimum, and the convergence time of cuckoo search algorithm is slow, so a maximum power point tracking control algorithm based on improved cuckoo search algorithm and particle swarm optimization algorithm is proposed. With the minimum convergence time as the constraint function, the optimal power interval is determined by parameter optimization and the critical point parameters is divided. The optimization process is divided into two stages: particle swarm fast coarse optimization and improved cuckoo search steady precision optimization, so as to improve the convergence speed and power generation efficiency of the algorithm. The simulation results show that the proposed algorithm is superior to other algorithms when the convergence time is 024 s and the power generation efficiency is 99.89% under the condition of uniform temperature field, and when the convergence time is 0.13 s and the power generation efficiency is 99.92% under the condition of nonuniform temperature field. The algorithm converges quickly and has high tracking accuracy, and has passed the benchmark test function test. The effectiveness and universality of the algorithm are verified.

Abstract:To solve the problems of large positioning error of Radio Frequency identification indoor positioning algorithm, an indoor positioning algorithm based on improved gray wolf optimization algorithm is proposed by applying intelligent algorithm to indoor positioning algorithm. For the traditional grey wolf optimization algorithm has the problem of low convergence accuracy and easy to get the global optimal solution,the nonlinear convergence factor based on the power function to increase the algorithm′s optimizationseeking ability; the exponential factorbased position update strategy is used to heighten he convergence accuracy of the algorithm; and adds a multiple position update strategy to make the algorithm easily jump out of the local optimal solution. The experimental results show that the positioning error of the traditional trilateral localization algorithm is 0887 m, and Indoor localization algorithm based on improved grey wolf optimization algorithm can effectively achieve the target positioning with an average positioning error of 0.276 m, which significantly improves the positioning accuracy.

Abstract:The previous deep learning fault diagnosis methods for transmission lines rely on digital signal processing technology to extract fault features. In order to improve the above methods, this paper introduces graph deep learning theory and proposes an endtoend intelligent fault diagnosis method based on graph attention network. The original threephase current and voltage signals are converted into graph data, and the feature is automatically extracted using multiple graph attention layers, thus establishing the mapping relationship between the data from input to output, and realizing endtoend fault diagnosis of transmission lines. The accuracy and effectiveness of the method are verified on the 400 kV threephase transmission line and the IEEE13 bus power grid system, and the simulation analysis is carried out for five kinds of short circuit fault and no fault conditions with different initial phase angle, transition resistance and fault location. The results show that the fault diagnosis accuracy of this method is more than 99.75%, and its performance is the best compared with several existing intelligent fault diagnosis algorithms. At the same time, the method still maintains high fault identification rate under different white noise, has good robustness and generalization ability, and provides a certain research idea for power transmission line diagnosis technology.

Abstract:In view of the shortcoming of traditional ant colony algorithm in path planning of AGV, such as a large number of inflection points and high operating energy consumption, an improvement adaptive ant colony algorithm is proposed in this paper. Firstly, the adaptive parameter adjustment method to continuously adjust the relative importance of pheromone concentration and heuristic information in the iterative process to enhance the direction of ant search; Secondly, the multiobjective path performance evaluation index is introduced. Based on the single index of path length, the path risk index and the steering times are introduced to achieve the global comprehensive optimization of AGV path planning; Then a reward and punishment mechanism is proposed to update the pheromone increment, which provides different pheromone update rules for the paths of different evaluation indicators to avoid the algorithm falling into premature; Finally, the quasi uniform cubic Bspline smoothing strategy is introduced to further optimize the optimal solution. At 20×20 and 30×30 Simulation experiments are carried out in different complexity environments. Compared with the traditional ACO, the steering times of improved ACO is reduced by 11.3%~38.2%, and the path optimization speed is increased by 79.8%~87.9%, which verifies the effectiveness, feasibility and superiority of the improved ACO.

Abstract:To improve the fringe projection efficiency and measurement accuracy of the optical 3D measurement system, a binary defocusing method is proposed to rapidly project the coded fringes and the binocular stereo matching algorithm is used to obtain the parallax information. First, the Bayer dithering technology is used to convert the projected grayscale fringes into binary fringes to reduce the effect of the projector′s defocusing on the contrast and depth of field drop. Then, the wrapped phase is expanded into absolute phase by means of multifrequency heterodyne method, disparity constraint relation and average error sum of squares cost calculation are used to obtain the best phase matching point. Finally, the depth information and threedimensional information of the measured object are obtained through the principle of triangulation and dualtarget positioning. Through experiments on precision balls and statues, it is proved that the measurement accuracy of the method proposed in this paper is close to that of the traditional twelve-step phase shift measurement, and is 52.8% higher than the traditional three-step phase shift.

Abstract:The fusion of multiscale receptive field feature can remarkablely improve the detection accuracy of models, but it also greatly increases the computational cost of models at the same time. To address this issue, we propose the object detection model based on structural reparameterization. Firstly, max pooling in SPP is substituted by depthwise convolution, while structural reparameterization is utilized to reduce computational complexity of module simultaneously. Based on this, we propose a new multiscale receptive field feature fusion module, called CspRepSPP. Additionally, a new feature extraction module, named RepBottleNeck, is proposed according to structural reparameterization. Experimental results show that, compared with the original YOLOv5s model, the mAP05:095 of our model is improved by 322 percentage points, the detection speed of single image is improved by 05 ms, and the GFLOPs is reduced by 10. Compared with other improved methods based on YOLOv5s, our method shares higher detection accuracy, faster inference speed, and lower number of parameters.

Abstract:Aiming at the TLD target tracking algorithm is affected by the problems of occlusion, rotation and motion blur in the actual tracking process, an improved TLD algorithm is proposed, and an improved FAST corner point algorithm is introduced in the tracking module, as well as a positive sample library under the data enhancement method of randomly cut blocks in the detection module. A rotation invariant LBP algorithm is introduced to design a classifier with rotation invariance. The improvement of the tracking module not only ensures the accuracy of the algorithm, but also improves the realtime performance of the algorithm. The improvement of the detection module allows the TLD algorithm to track well even when the moving target is partially obscured or has changed its shape. Experimental verification of the data set OTB2013 reveals that the algorithm in this paper not only has high tracking accuracy and fast running speed compared with other five algorithms, but also can effectively improve the tracking effect when the moving target is obscured.

Abstract:Aiming at the problems that small target features in remote sensing images are easily lost, easily affected by background noise and difficult to locate, this paper improves the YOLOXS target detection model. Firstly, the CBAM is improved by using the twodimensional discrete cosine transform and added to the backbone network to improve the attention of the network to small targets; secondly, a weighted multireceptive spatial pyramid pooling module is proposed to improve the perception ability of the model to multiscale targets, especially to smallscale targets. Thirdly, using the idea of crosslayer feature fusion, a crosslayer attention fusion module is proposed to retain as many small target features as possible in the deep structure; finally, EIoU loss is used to enhance the localization ability of small targets. As shown by extensive experimental analysis, the APs value of the improved model improves by 51% relative to the baseline model on the RSOD dataset and by 24% on the DIOR dataset, and the number of parameters increases by only 1.01 M. The detection speed reaches 93.6 fps, which meets the detection requirements of real-time. In addition, the improved model in this paper also has certain advantages over the current stateoftheart target detection models.

Abstract:Uncertain object auto detection is the key technology of unmanned intelligent lifting handling, and efficient technology recently used is Instance segmentation model based on deep learning. Due to the limited ability of the existing cases to segment the trunk network to extract the global context feature information of lifting scene images, and the difficulty of the convolutional operators in the convolutional neural networkbased trunk network to model the long range correlation of the receptive field, and the lack of sufficient depth cues when identifying single targets with texture features, a module is designed to integrate the heterogeneous feature information of CNN and Transformer, and Transformer is used to model the global dependency relationship, and it is integrated with the ability of CNN to extract local information. Then, the Dense RepPoints detection network was introduced to construct the case segmentation network for the complex lifting and loading scenarios, which could accurately segment the loading and unloading objects and different surfaces of the objects. Compared with the most advanced method at present, AP increased by 4.95% to 98.82%, mIoU increased by 542% to 9189%, obtaining a good example segmentation effect, solving the key technical problems of intelligent lifting loading and unloading, thus improving the work efficiency and safety of unmanned lifting loading and unloading logistics transportation, and reducing costs.

Abstract:To solve the problem of clinical inability to predict urinary stone types preoperatively, we propose a method for preoperative prediction of stone types based on CT images, and develop a preoperative prediction aid system for urinary stone types based on improved residual network. This enables preoperative prediction of stone types. The main tasks includes: firstly, Resnet34 is used as the base network with improved pooling layer, residual block structure and loss function, and a dense connection structure is added. Thus, the problem of stone misidentification is solved. Secondly, a twobranch multiheaded selfattentive module was designed to increase the feature weights of the stone region, which greatly improved the model performance. Finally, through a retrospective study, a dataset containing 5 709 CT images of stones was created and randomly divided into a training set, a validation set and a test set in the ratio of 3∶1∶1, which was used for training and performance validation of the model. The proposed improved residual network was experimentally verified to be 72.90% accurate on the test set, with a 6.38% improvement in accuracy and a 10% increase in F1 score. The results showed that the model can effectively improve the accuracy of preoperative prediction of urinary stones and has potential clinical application.

Abstract:The recognition and positioning capabilities of the visual perception terminal of the fruitpicking robot system are crucial indicators to increase fruitpicking success rates in the complicated agricultural environment. A realtime multitask convolutional neural network SegYOLOv5 suited for autonomous Pitaya fruit image detection for the visual system of the picking robot was proposed in this paper using Pitaya fruit with complicated shape as the research object. The network is enhanced based on the primary architecture of YOLOv5′s convolutional neural network. The multitasking target recognition and detection task of image detection and semantic segmentation is realized, and the overall performance of the model is substantially improved, by extracting threelayer enhanced features as the input of the improved cascaded RFBNet semantic segmentation network layer. With a mean Average Precision and mean Intersection Over Union of 9310% and 8364%, respectively, for the testing dataset, the enhanced SegYOLOv5 network architecture can adapt to the boundary-sensitive image semantic segmentation agricultural scene, compared with YOLOv5s+original RFBNet and YOLOv5s+BaseNet models, it is 123% and 274% higher than the former, and 238% and 145% higher than the latter. The average detection speed of SegYOLOv5 can reach 7194 fps which is 4079 fps faster than EfficientDetD0, and the mean Average Precision is 58% higher. The center of mass of Pitaya fruit may be precisely positioned in real time as the best picking position using the endtoend output of SegYOLOv5 detection output and the fusion of image geometric moment operator. The improved algorithm has high robustness and versatility, which lays an effective practical foundation for fruit picking robot based on visual perception.

Abstract:Instance segmentation is of great significance for eliminating safety hazards brought by irregular machinery and equipment on construction sites and for monitoring workers. However, the current mainstream instance segmentation models have the problem of low boundary detection accuracy. Combining the characteristics of instance segmentation, this paper proposes an improved Mask R-CNN model of multi-stage refining mask based on the global context channel attention (GCCA) mechanism. First, this paper gradually fuses finegrained features in the mask head in a multi-stage manner to refine high quality masks. Second, in order to better fuse fine-grained features, a GCCA attention mechanism is constructed, which aggregates global features through a simplified global context module, and utilizes onedimensional convolution to achieve local channel interactions without dimensionality reduction. The experimental results show that this paper has achieved great results on both COCO and MOCS datasets. Among them, compared with the Mask R-CNN model, the average accuracy of the algorithm in this paper in detection and segmentation is improved by 2.4% and 7.6% respectively.

Abstract:To address the problem that the existing algorithms for generating infrared images from visible images cannot perceive the weak texture regions of the images, which leads to the low quality of the generated image details, this paper proposes an improved Cycle Generation Adversarial Network (CycleGAN) for the image generation task. structure for image generation tasks. Firstly, the generator network structure of the cycle generation adversarial network is constructed by using the residual network with stronger feature extraction ability, so that the image features can be fully extracted and the problem of low image quality caused by insufficient feature extraction can be solved; secondly, the channel attention mechanism and spatial attention mechanism are introduced in the generator network structure, and the regions with poor image perception are weighted by the attention mechanism to solve the problem of loss of image texture details. processing to solve the problem of image texture detail loss. On the OSU dataset, the proposed method improves the Peak Signal to Noise Ratio (PSNR) and Structural Similarity (SSIM) metrics by 7.1% and 10.9%, respectively, compared with the cyclic generative adversarial network method on the Flir dataset. PSNR and SSIM improved by 4.0% and 6.7%, respectively, on the Flir dataset. The experimental results on several datasets demonstrate that the improved method in this paper can highlight the detailed feature information in the image generation task and can effectively improve the quality of image generation.

Abstract:The balise is the basic information transmission equipment of the train control system. It has high precision and reliability when used for positioning. However, its core component, radio frequency identification (RFID) tags, is monopolized by foreign manufacturers. Therefore, a balise auxiliary positioning system based on long range radio (LoRa) is designed. The system uses the peak value of the received signal strength indication (RSSI) as the basis for judging the position of the train query to the balise, and obtains the accurate peak time of the RSSI value through the envelope and interpolation processing method. The time is combined with the mileage data of the odometer (ODO) to achieve accurate positioning. The experimental results show that the positioning accuracy of the system is 1.32 m, 2.15 m and 3.4 m respectively when the speed of the query is 15 km/h, 30 km/h and 60 km/h. The experimental results show that the positioning method proposed in this paper can realize the auxiliary positioning of the train, which provides a feasible technical scheme for the localization of the balise.

Abstract:Aiming at the problems of inadequate feature extraction and low recognition accuracy in human motion recognition, a human motion recognition model based on improved Longterm Recurrent Convolutional Network was proposed. Firstly, a LRCN model composed of multi-layer convolutional neural network and gated circulation unit is constructed. On this basis, the internal and external cycle layers are constructed. The role of the internal cycle layer is to obtain the internal time characteristics and spatial characteristics of the selected time window, while the role of the external cycle layer is to obtain the feature correlation and time correlation between the state information represented by the subsequence data. The proposed model was verified on three public data sets with higher accuracy than the traditional LRCN model. Then, it was tested and verified on the self-built data sets, and the recognition accuracy was 99.7%. The experimental results show that the recognition accuracy of this model is higher than that of the original model, which verifies the feasibility of this model.