Search

Real-time target detection in traffic scenarios is the prerequisite of electronic monitoring, automatic driving, and other functions. In view of the low detection efficiency of existing target detection algorithms and the low accuracy of most light target detection algorithms, which are easy to obtain wrong or insufficient target detection, this paper adopts the improved YOLOv5 target detection algorithm for model training, and the pseudo-label strategy for training process optimization. Then, the KITTI traffic target dataset tags are merged into three categories for model training and testing. Through the experimental comparison, the improved YOLOv5 model in this paper achieves 92.5% mAP in all categories, which is 3% higher than the original YOLOv5 training model. Finally, the three categories of the trained models are deployed on the Jetson Nano embedded platform for inference testing, and TensorRT is used to accelerate inference. The average inference time per frame of image is 77 ms, which meets the goal of real-time detection. [ABSTRACT FROM AUTHOR]

Model counting is the number of models for a given proposition formula, which is a basic problem in the field of artificial intelligence. There are many exactly-one constraints in many practical problems such as Bayesian networks, bounded model detection, and accurate set coverage. A common processing method is to encode exactly-one constraints as CNF formulas, and then call the model counter to solve them. This method expands the scale of the proposition formula and easily leads to too long solution time. This paper respectively proposes the ECR algorithm that restores exactly-one constraints from the CNF formula and the ECP algorithm that handles exactlyone constraints. The ECR algorithm can significantly improve the solution efficiency of the C2D compiler. Based on the latest model counter ExactMC, this paper improves the model counter ECMC that can recognize and handle exactly-one constraints separately. The experimental results show that the time efficiency of ECMC is significantly improved compared to ExactMC. [ABSTRACT FROM AUTHOR]

Aiming at the precision loss problem of most low-complexity Root-MUSIC algorithms at present, a low-complexity Root-MUSIC algorithm with precision compensation ability is studied and proposed. The algorithm reconstructs the autocorrelation matrix with Toeplitz shape according to the first row of the approximate data observation matrix obtained by finite snapshots, so that the reconstructed autocorrelation matrix has Hermitian property. After decomposing the reconstructed autocorrelation matrix, the noise subspace is obtained, the noise subspace is flipped and split, a new root-finding polynomial is reconstructed, and then the DOA estimated value is obtained by the root-finding method. The algorithm proposed in this paper using Toeplitz matrix reconstruction and root polynomial reduction effectively improves the DOA estimation accuracy of the improved Root-MUSIC algorithm. And the time complexity of the improved algorithm is no higher than that of previous algorithms. Under different incident sources and sampling snapshots, the algorithm proposed in this paper also shows stronger robustness and stability. [ABSTRACT FROM AUTHOR]

Network representation learning can effectively solve the problem of data sparsity in recommendation. In this paper, LINE and DeepWalk in network representation learning were improved, and a hybrid recommendation algorithm was proposed to be applied to movie recommendation scene. The new algorithm generates three low dimensional feature vectors by learning user preference feature, user aversion feature and similar user feature. Three low dimensional feature vectors are linearly combined to form a user representation vector, and cosine similarity is used as the similarity index to recommend the movies associated with similar users to target users. Experimental results show that, compared with the suboptimal algorithm, the accuracy and Fl index of the proposed algorithm are improved by 12% and 7% respectively on MovieLens dataset, and 16% and 18% respectively on MovieTweetings dataset. The recommendation algorithm based on multi-dimensional feature representation learning proposed in this paper has significant advantages in movie recommendation scenes. [ABSTRACT FROM AUTHOR]

Aiming at the problem that the classic puncture path planning algorithm does not consider the area around the path when quantifying the path, new constraints condition are proposed to make up for the lack of quantification of the information around the path. A strict constraint condition was proposed based on the blood vessel position information (the surrounding area avoids the blood vessel), and based on bone density, blood vessel density, and diameter information, three soft constraints (bone density, blood vessel density, and blood vessel risk) were proposed, and the surrounding information was added to the quantification of the path to improve the safety of the puncture path. In order to adapt to the specificity of patients, a soft constraint optimization degree scoring algorithm is proposed, and the weight is set according to the optimization degree of soft constraints, and the puncture path corresponding to the maximum optimization degree is taken as the optimal puncture path. In order to verify the effectiveness of the algorithm, experiments were carried out on the public dataset 3D-IRCADb. The experimental results show that the algorithm proposed in this paper can quickly calculate the optimal puncture path that meets the clinical needs. [ABSTRACT FROM AUTHOR]

Since the left ventricle pressure waveform（LVPW）estimation is based on the model with numerous patient specific parameters, there is a great need to reduce the computation cost. A parameter subset selection strategy based on sensitivity analysis is proposed to solve these problems. LVPW features are chosen as outputs of a systemic circulation model. The meta model is created by adaptive sparse polynomial chaos expansion algorithm, and then Sobol sensitivity index is computed. Finally, the parameters which have large impact on LVPW features are selected as parameter subset. The parameter subset selection proposed in this paper can provide help for applying model to patient specific situations in clinical applications. Meantime, the results indicate that the reducing size of parameter subset can decrease the complex computation of parameter optimization significantly. Furthermore, LVPW estimation based on the proposed method has high correlation with that based on full model parameters. [ABSTRACT FROM AUTHOR]

Aiming at the problem of inaccurate locating of traditional iris locating algorithms, an improved YOLOV3 iris locating model is proposed to improve the accuracy of iris locating and make it better applied to production practice. Using the Densenet-121 model as the feature extraction module, and on the basis of it, the auxiliary network is obtained by copying the backbone network to make it more conducive to the detection of small targets, and the non-local attention mechanism is used to enhance the semantic information of the features obtained by the image. The YOLOV3 model, Daugman model and Wilde model based on DarkNet-53 are used for comparative experiments. The experimental results show that the accuracy of the experimental model in this paper is as high as 97. 1 % in iris locating, which has obvious advantages compared with other iris locating models. [ABSTRACT FROM AUTHOR]

The ECGP wave contains abundant physiological and pathological information of the human body, but the P wave amplitude is small and the shape is changeable, which is very difficult to detect. This paper proposes a bilateral long-short-window energy difference method to detect multi-modal P wave boundaries, and uses the particle swarm algorithm to optimize its parameters. 4363 and 1936 ECG beats are selected from the LUE database and QT database respectively, with 70% and 30% of them used as the training set and the test set, respectively. Compared with the parameter-mixed Gaussian fitting method based on dynamic programming and the phase-free stationary wavelet transform method, the error of the results of positive and negative P waves is significantly smaller than the above two methods, and the algorithm can detect bidirectional P waves, and it also has anti-noise ability. [ABSTRACT FROM AUTHOR]

CAN( controller area network) bus is the most widely used field bus. Due to the lack of authentication and message checking mechanism, CAN bus has great security risks, so it is necessary to design a defense mechanism for CAN bus. In this paper, a priority hopping mechanism is designed, which introduces the dynamic hopping of identifiers by Hash function into a real-time scheduling algorithm. The fixed priority is calculated by a genetic algorithm, and the compromise range of priority is found out. Each frame is grouped, and the ID segment of data frame is segmented and reconstructed. The former part of the ID segment determines the priority and performs priority hopping, while the latter part of the ID segment hops dynamically at one time. Experimental results show that using dynamic priority and one-time ID Hopping to hop has a greater security improvement than the existing ID Hopping mechanism. [ABSTRACT FROM AUTHOR]

The multi-scale quantum harmonic oscillator optimization algorithm(MQHOA)is a natural calculation algorithm based on quantum physics proposed in recent years. Aiming at the problem that the algorithm fails to make full use of the historical information in the iteration, this paper proposes a historical information-driven multi-scale quantum harmonic oscillator optimization algorithm(HI-MQHOA). In the two-step iterative process, HI-MQHOA introduces historical data as a driver to form the parameters of the next generation individual distribution and dynamically adjust the scale of the algorithm. The next generation individual distribution parameters can effectively guide the development and exploration of the algorithm, and dynamic scaling can avoid premature stagnation. Verified by several classical test functions, the algorithm is superior to MQHOA, improved MQHOA and other natural computing algorithms in solution quality, accuracy and scalability. [ABSTRACT FROM AUTHOR]