Graph conjoint attention network (CAT) is one of the best graph convolutional networks (GNNs) frameworks, and has achieved the most advanced performance in semi-supervised node classification tasks. CAT finds important neighbor nodes through a weighting mechanism based on context and node content. However, this weighting mechanism is based on the static context for learning, that is, it treats each node and edge equally, which means that the weighting mechanism is vulnerable to noise nodes and edges, so it has significant limitations. In this paper, we propose a method to obtain context dynamically based on random walk, which makes the context based weighting mechanism better able to avoid noise interference. Furthermore, we combine the proposed context based weighting mechanism with the node content based weighting mechanism of graph attention network (GAT) to form a model based on a mixed weighting mechanism. The model can better discover important neighbors, eliminate noise edges, and learn node embedding by message passing. Our experiments show that the proposed model achieves state-of-the-art performance on node classification tasks on multiple datasets.

n the classification problem, deep kernel extreme learning machine (DKELM) has the characteristics of efficient processing and superior performance, but its parameters optimization is difficult. To improve the classification accuracy of DKELM, a deep kernel extreme learning machine algorithm optimized by the improved sparrow search algorithm (ISSA-DKELM) is proposed. The deep kernel extreme learning machine classifier is developed with the optimal parameters using ISSA to address the DKELM parameters selection problem. In order to make up for the shortcomings of the basic sparrow search algorithm, the chaotic transformation is first applied to initialize the position of the sparrow position. Then, dynamically adjust the location of the discoverer's sparrow population. A 'learning operator' in the teaching-learning-based algorithm is fused to improve the position update operation of the joiners. Last, the Gaussian mutation strategy is added in the later iteration of the algorithm to make the sparrow jump out of local optimum. The experimental results show that the proposed DKELM classifier is feasible and effective, and compared with other classification algorithms, better test accuracy is achieved.

The continuous phase modulation (CPM) technique is widely used in range telemetry due to its high spectral efficiency and power efficiency. However, the demodulation performance of the traditional Maximum Likelihood Sequence Detection (MLSD) algorithm significantly deteriorates in non-ideal synchronization or fading channels. To address this issue, this work proposes a convolutional neural network (CNN) called the cascade parallel crossing network (CPCNet) to enhance the robustness of CPM signals demodulation. The CPCNet model employs a multiple parallel structure and feature fusion to extract richer features from CPM signals. This approach constructs feature maps at different levels, resulting in a more comprehensive training of the model and improved demodulation performance. Simulation results demonstrate that, in the absence of timing errors and with only Gaussian noise, the bit error rate (BER) performance of the CPCNet approaches the theoretical value of the MLSD method for Gaussian minimum shift keying (GMSK) and frequency shift keying (FSK) signals. Moreover, compared to the MLSD algorithm, for GMSK signals the BER of the CPCNet decreases by 5% to 87% in the signal noise power ratio (SNR) range from -2dB to 18dB when Rayleigh fading is present. Additionally, with a timing error of one-quarter of the symbol period, the CPCNet exhibits an average demodulation gain of 2dB compared to other deep learning methods such as CNN and CLDNN. Furthermore, the demodulation gain increases as the timing error increases, indicating the superior robustness of the CPCNet.

For the task of content retrieval, analysis and generation of film and television scene images in the field of intelligent editing, fine-grained emotional recognition and prediction of images is of great significance. In this paper, the fusion of traditional perceptual features, art features and multi-channel deep learning features are used to reflect the emotional expression of different levels of the image. In addition, the integrated learning model with stacking architecture based on linear regression coefficient and sentiment correlations (LS-stacking) is proposed according to the factor association between multidimensional emotions. The experimental results prove that the mixed feature and LS-stacking model can predict well on the 16 emotion categories of the self-built image data set. This study has improved the fine-grained recognition ability of image emotion by computers, which helps to increase the intelligence and automation degree of visual retrieval and postproduction system.

Mobile manipulators with high degrees of freedom have increasingly been employed in several scenarios such as patient caretaking, logistics, and human robot collaboration. To achieve a high efficiency of motion planning for mobile manipulators autonomously is essential for their tasks. Hierarchical motion planning algorithms have been shown to be effective for mobile manipulators. However, a scenario with narrow-channel introduces significant challenges for these algorithms. To address the problem, an Improved Hierarchical Motion Planner (IHMP) is proposed, which consists of a two-dimensional path planner for mobile base, and a three-dimensional trajectory planner for on-board manipulator. The main framework is descried as follows. Firstly, a hybrid sampling strategy is proposed, that can reduce invalid nodes of the generated probabilistic roadmap. Bridge test is used to locate the narrow channel region, and a gaussian sampler is deployed in these regions and the boundaries. Meanwhile, a random sampler is deployed in rest regions. Trajectory planner for on-board manipulator is to generates a collision-free and safe trajectory in the narrow channel with collaboration of the two-dimensional path planner. The experimental results shows that the proposed method is effective for mobile manipulator motion planning in complex static environments, especially in narrow channel.

In order to improve the accuracy of second-hand car price prediction, a machine learning pre-diction model based on the retention rate was proposed in this paper. Firstly, random forest al-gorithm was used to filter the variables in the data. Seven main characteristic variables that affect used car prices, such as new car price, service time, mileage and so on, were filtered out. Then, the linear regression classification method was introduced to classify the test data into high and low retention rate data. Afterwards, XGBoost (Extreme Gradient Boosting) regression model was built for the two sets of data respectively. Prediction results show that the comprehensive evaluation index of the proposed model is 0.548, which is significantly improved compared with 0.488 of the original XGBoost model. Lastly, compared with other representative machine learning algorithms, this model shows certain advantages in Mean Absolute Percentage Error, 5% accuracy rate and comprehensive evaluation index. As a result, the machine learning model based on retention rate established in this paper has significant advantages in terms of used car price prediction accuracy.

In this letter, a miniaturized lumped element lowpass filter (LPF) based on four-layer liquid crystal polymer (LCP) substrate is proposed. An 8-shaped inductor has been used to reduce the size and improve the performance of the designed filter. The simulated results indicate that the volume of the 8-shaped inductor is reduced by 52% efficiently compared to the multilayer spiral inductor, and the Q-factor of the 8-shaped inductor is 83. Using this type of inductor and vertically interdigitated capacitors (VICs), a lumped element five-order lowpass filter with the cut off frequency of 0.5 GHz is designed, manufactured, and measured. The measured results indicate that the designed filter introduces a transmission zero at 0.95 GHz, making the rectangular coefficient of 1.57 and the stopband responses over 40 dB at 0.9~3 GHz, with an insert loss (IL) of 0.37 dB, and a return loss (RL) of 30 dB in the passband. The whole volume of the proposed filter is 0.023λg × 0.015λg × 0.00032λg (λg is the guided wavelength of the cut off frequency). The proposed filter is low-cost to manufacture and easy to integrate into radio frequency front-end (RFFE) system.

In this letter, a W-band ultra-broadband microstrip (MS)-to-microstrip (MS) vialess vertical transition in multilayer liquid crystal polymer (LCP) substrate is presented, which consists of two microstrip lines in the top layer, a common ground plane and slotline resonators in the second layer, and a close-loop transition-line in the bottom layer. To largely improve the passband of the vialess vertical transition, an H-shaped slotline resonator has been introduced. It changes the impedence of the slotline resonator so that return loss (RL) and insertion loss (IL) are greatly improved. Full-wave simulation indicates that IL is less than -2 dB and RL is better than -10 dB at W-band. To verify this design, the ultra-broadband vertical transition is fabricated and measured. Owing to the error of the fabrication, a frequency ranging from 70 to 104.09 GHz can be obtained with RL better than -10 dB and IL less than -5.67 dB.