Search

The existing YOLO target detection algorithm, based on one-stage concept, aims for multi-objective detection. But this algorithm performs not well for dual classification detection and add more performance overhead while detection. In order to improve the detection efficiency of dual classification mask wearing during the period of COVID-19, this paper proposes a real-time detection method based on YOLO for detecting the condition of bi-objective mask wearing. The optimization is based on the latest YOLO version 4: firstly, the feedforward input layer of the model is improved to optimize the data augment part and the adaptive image scaling is added to improve the efficiency of the detection for both dual classification and small objectives; secondly, the adaptive anchoring frame is added to replace the activation function so as to reduce the computation and thus improve its efficiency; thirdly, the optimization of Neck and the addition of Focus structure improve the capability of feature fusion and reduce the amount of parameters to raise the efficiency. The experimental results showed that, compared with the existing YOLO version 4, using this method in this paper improves F1 by 0.33% and mAp by 0.71% in the data set. Also, the detection efficiency is also improved significantly under the same experimental condition. [ABSTRACT FROM AUTHOR]

With the continuous growth of vehicle ownership and the popularization of Internet of Vehicles applications， vehicle terminals generate large amounts of data messages that need to be processed in real time. In the scene of high-speed vehicle movement， the traditional Internet of Vehicles navigation system has a transmission delay due to vehicle differential positioning data， which results in a certain deviation in vehicle positioning results and cannot obtain high-precision positioning results in time. Based on this， this paper proposes a technology solution for car networking navigation based on BDS positioning and edge computing. It uses an improved genetic algorithm to allocate resources for terminal positioning requests， effectively reduces the service delay of the entire edge network， and uses optimize the lossless Kalman filter algorithm based on edge node to improve the positioning accuracy of the car network node. Experiments show that the method proposed in this paper can provide real-time accurate， low-latency and high-precision positioning services for large-scale Internet of Vehicles terminals， and has high practical application value. [ABSTRACT FROM AUTHOR]

Aiming at the low accuracy and time-consuming problems of AKAZE algorithm in image matching of glass-encapsulated electrical connectors, an improved image matching algorithm based on LK optical flow and grid motion statistics is proposed in this paper. Firstly, the AKAZE algorithm is used to extract the feature points of the glass-encapsulated electrical connector, and the M-LDB descriptor is used for the feature description. Then, LK optical flow method is used to calculate the matching area for conditional constraints, so as to obtain matching points, and the FLANN algorithm is used for feature matching. Finally, the image of the glass-encapsulated electrical connector is divided into multiple grids, and the number of correct matching points in the neighborhood of the well-matched feature points by FLANN is calculated and the threshold value is used to eliminate the incorrect matching points. The experiments in this paper use public data sets and glass-encapsulated electrical connector data to verify and analyze the performance of the algorithm in terms of real-time and accuracy. The results show that the improved algorithm has a matching accuracy of more than 93% when dealing with image pairs of glass-encapsulated electrical connectors with changes in blur, brightness and rotation, and takes less than 0.4 s, which proves the effectiveness of the algorithm. [ABSTRACT FROM AUTHOR]

Aiming at many problems caused by manual acceptance in decoration projects, this paper proposes an improved SSD algorithm and applies it to supervision work to replace manual acceptance and promote the realization of intelligent supervision. Since the SSD algorithm has problems such as re-examination of the same target and poor detection of small targets, the DPN network is used to replace the basic feature extraction network VGG16 in this paper. DPN combines the advantages of Resnet and Densenet and has better feature extraction ability. The feature maps are fused by weighted FPN to highlight the contributions of feature maps of different layers and enrich the semantics of feature maps for prediction. The use of depthwise separable convolution reduces the number of parameters of the model and improves the inference speed of the algorithm. The experimental comparison shows that the average accuracy of the improved model is increased by 3.47%, and the average accuracy of small number detection is improved by up to 15%, which proves that the new model has a good effect in the task of supervision target detection. [ABSTRACT FROM AUTHOR]

For the scheduling algorithm applied in CAN FD network, when the earliest deadline algorithm of average partition coding method performs non-preemptively scheduling on messages, it will have the problems of limited encoding ability for a large range of deadlines, high probability of priority reversal of messages and high bus load. By analyzing various causes that cause of message transmission delay and combined with the shortcomings of previous partition scheduling algorithms, this paper proposes an improved method of message scheduling based on the earliest deadline first algorithm based on exponential – power function partition, that is, on the basis of exponential partition of message deadline, power function partition subdivision is further adopted. At the same time, the schedulability of the algorithm is analyzed and verified by simulation with CANoe. Experiments show that compared with the existing average partition scheduling algorithm, the improved algorithm expands the representation range of deadline, reduces bus load, optimizes priority inversion problem, and achieves better scheduling effect. [ABSTRACT FROM AUTHOR]

In view of the problems that node clock drift， network link delay and poor synchronization ability in Ethernet， this paper constructs a clock synchronization system model based on the communication principle of RTEthernet protocol. By studying the composition and working principle of RTEthernet， three factors affecting the time synchronization precision of real-time Ethernet-drift rate， network transmission delay and clock Byzantine fault are considered. Based on this， the original FTA clock synchronization algorithm is analyzed， and it is found that its fault tolerance performance is obviously reduced when Byzantine faults increase. The idea of "median fault tolerance" is introduced to improve it and RTE-FTM algorithm is proposed. Through the CANoe simulation platform， two Byzantine faults and no Byzantine faults in the system （seven nodes） are compared and analyzed， and the precision loss rate of the system is reduced by 3.1%， which verifies the convergence and effectiveness of the algorithm. [ABSTRACT FROM AUTHOR]

The well inclination and azimuth are the main measurement parameters in the calculation of the wellbore trajectory. However, compared with the azimuth error of the conventional well inclination, the azimuth angle measurement error of the inclinometer under the small angle inclination is larger. In order to improve the azimuth angle measurement accuracy of the inclinometer under the small-angle well inclination, the azimuth angle measurement under the 5°～10° small-angle well inclination was compensated based on the BP neural network algorithm. In this paper, a two-dimensional vector composed of standard well inclination and measured azimuth is used as input, and a one-dimensional vector composed of standard azimuth is used as output, and a network model of double input and single output is established. The learning samples are divided into training set and test set by random selection, so that the network has better generalization ability. The simulation test results show that the BP neural network error correction model runs stably and the compensation accuracy reaches 10 -6, which can improve the measurement accuracy of the azimuth angle under the small-angle well inclination from ±5.3° to within ±1.7°. [ABSTRACT FROM AUTHOR]

The optimization performance of the PSO algorithm is affected by the speed update formula. Too fast convergence speed may cause the algorithm to miss the global optimal solution; too slow convergence speed may cause the algorithm to fall into the local optimal solution. Aiming at this problem, this paper proposes a PSO optimization algorithm based on improved compression factor, namely FPSO. By introducing the compression factor equation, the speed iteration formula is improved, and the influence on the algorithm caused by the improper setting of the learning factor is reduced. The new adjustment mechanism not only ensures the convergence performance of the PSO algorithm, but also weakens the influence of the speed boundary on the algorithm. Finally, five classical functions are selected to test the performance of the algorithm. The test results show that, compared with the traditional PSO algorithm, the proposed algorithm improves the global convergence ability and shortens the convergence time. [ABSTRACT FROM AUTHOR]