Search

Few-shot fine-grained (FSFG) categorization aims to classify new, fine-grained concepts using limited annotated data for each class. It requires a model capable of learning subtle, distinguishable features from similar classes. An intuitive approach is to identify several discriminative parts based on the input image and represent it using these localized parts. It is interpretable since the parts obtained and combined can be seen as the way humans learn about new concepts. Therefore, we propose a straightforward and efficient module for the FSFG task by emphasizing the importance of information from discriminative parts. Specifically, we design a plug-and-play meta-detector based on object detection models, which identifies the positions of distinguishable parts in fine-grained images. With the positions of objects and semantic parts identified, we can extract richer features to enhance few-shot results. Extensive experiments on five fine-grained datasets demonstrate that our approach is applicable to various few-shot backbones and significantly improves accuracy without adding extra parameters. More importantly, compared to other variants on the same base few-shot network, the meta-detector module significantly outperforms them, further demonstrating its effectiveness.

In dynamic environments, the traditional relative positioning methods based on the Kalman filter model suffer from low accuracy and stability due to the influence of noise and outliers. This paper proposes a variational Bayesian filtering algorithm based on the combination of four-frequency observations from BDS (BeiDou Navigation Satellite System) and models the observation noise using the T-distribution to enhance the stability of filtering. Firstly, a geometrically correlated ambiguity resolution model is constructed based on the characteristics of the combined observations, effectively improving the precision of float ambiguity resolution and fixing rate. Moreover, considering the characteristics of outliers that are likely to occur in dynamic conditions, a T-distribution-based variational Bayesian filtering approach is employed to estimate the time-varying observation noise and system states. Experimental results demonstrate that the proposed method exhibits robustness and stability in dynamic short baseline scenarios, leading to further improvements in positioning accuracy, float ambiguity resolution precision, and fixing rate.

Graph Pooling technology plays an important role in graph node classification tasks. Sorting pooling technologies maintain large-value units for pooling graphs of varying sizes. However, by analyzing the statistical characteristic of activated units after pooling, we found that a large number of units dropped by sorting pooling are negative-value units that contain representative information and can contribute considerably to the final decision. To maintain more representative information, we proposed a novel pooling technology, called Geometric Pooling (GP), containing the unique node features with negative values by measuring the similarity of all node features. We reveal the effectiveness of GP from the entropy reduction view. The experiments were conducted on TUdatasets to show the effectiveness of GP. The results showed that the proposed GP outperforms the SOTA graph pooling technologies by $1\%~\sim ~5\%$ with fewer parameters.

Point cloud generation and representation is important in industry areas. Generating and editing high quality 3D shapes is challenging work in deep learning. Inspired by StyleGAN, a style based generative adversarial networks is proposed to generate high quality 3D point cloud. An improved non-linear mapping network learn distribution of points and is used to generate well distributed point cloud point cloud. We also provide a coarse to fine representation for point cloud. According to the experimental results on the ShapeNet Part data set(including aircraft, chair single category and overall 16 categories), our method can generate more uniform point cloud than other GAN methods with less training epoches. The latent code for point cloud has better linear separation,and is more easy to edit.

Wireless communication network (WCN) is very important for providing convenient mobile network communication services. Random Phase Multiple Access (RPMA) WCN under heterogeneous network architecture is widely used in wireless network construction worldwide due to its low power consumption and high cell density. However, this kind of WCN cannot meet the application scenario of high communication quality. Therefore, this research builds an RPMA communication quality prediction model for big data wireless base stations, which combines the convolutional neural network (CNN) algorithm and the lifting regression tree algorithm. It can be used to find the elements that have obvious influence on the communication quality. The reason for choosing the convolutional neural network algorithm is that its nonlinear feature relationship search and processing ability is excellent, and its computational complexity is relatively small. However, it is a black box algorithm and cannot obtain the importance coefficients of each feature, which is not conducive to subsequent analysis. Therefore, it is also necessary to select a relatively simple modified regression tree algorithm to participate in the calculation. The model constructed by integrating convolutional neural network and lifting regression tree algorithm is the RPMA wireless big data base station communication quality prediction model. A base station planning and deployment model based on weighted K-centroids algorithm is designed to obtain a better base station deployment scheme. In the CNN-DT model, the importance coefficients of T_B_diff, P_La and P_Lo are the largest and significantly larger than those of other features, which are 0.352, 0.289 and 0.264 respectively. The weighted K-centroids clustering algorithm designed in this study has the best overall downlink reception signal (RSSI) value distribution. For RSSI bucket “- 140 $\sim $ - 130”, the number of test points of WK centroids model, K-means model, GMC model, Mean Shift model and spectral clustering model accounted for 1.95%, 6.25%, 4.25%, 8.22% and 7.13% respectively. The model constructed in this study based on CNN and improved regression tree algorithm can accurately predict the communication quality of wireless big data base stations. This study’s main contribution is that it can be used in conjunction with the base station planning and deployment model based on weighted K-centroids algorithm to improve the accuracy and effectiveness of location selection for the RPMA wireless communication network base station.

As a result of the difficulties brought by COVID-19 and its associated lockdowns, many individuals and companies have turned to robots in order to overcome the challenges of the pandemic. Compared with traditional human labor, robotic and autonomous systems have advantages such as an intrinsic immunity to the virus and an inability for human-robot-human spread of any disease-causing pathogens, though there are still many technical hurdles for the robotics industry to overcome. This survey comprehensively reviews over 200 reports covering robotic systems which have emerged or have been repurposed during the past several months, to provide insights to both academia and industry. In each chapter, we cover both the advantages and the challenges for each robot, finding that robotics systems are overall apt solutions for dealing with many of the problems brought on by COVID-19, including: diagnosis, screening, disinfection, surgery, telehealth, care, logistics, manufacturing and broader interpersonal problems unique to the lockdowns of the pandemic. By discussing the potential new robot capabilities and fields they applied to, we expect the robotics industry to take a leap forward due to this unexpected pandemic.

Attention is an important field to explore the importance of each convolutional kernel channel/weight. The existing attention methods mostly use the Squeeze-and-Excitation (SE) technology to extract the global nonlinear feature vectors as the weights of corresponding feature maps. However, the pooling operators and fully-connected layers used in SE technology extract global features at the cost of much valuable information loss and the parameter amount increase. Actually, the feature map containing full information is a ready-made and better attention for other feature maps in the same layer. Simultaneously the products of feature maps will bring powerful non-linearity. Seeing this, Kernel Product (KP) technology is proposed to simply get useful nonlinear attention. To verify the effectiveness of KP, the proposed KP module is employed on Selective Kernel Networks (SKNets) to take the place of the original SE technology. The variety of SKNets is called Kernel Product Networks (KPNets) in this paper. In addition, identity mapping is used to solve the non-convergence problem in very deep neural networks. The KPNets are evaluated on ImageNet-1k, CIFAR-10, and CIFAR-100. The experiment results show that KPNets outperform many state-of-the-art methods and get a similar but more efficient performance than its SKNets with counterpart.

The Vehicle Routing Problem with Time Windows (VRPTW) has drawn considerable attention in the last decades. The objective of VRPTW is to find the optimal set of routes for a fleet of vehicles in order to serve a given set of customers within capacity and time window constraints. As a combinatorial optimization problem, VRPTW is proved NP-hard and is best solved by heuristics. In this paper, a hybrid swarm intelligence algorithm by hybridizing Ant Colony System (ACS) and Brain Storm Optimization (BSO) algorithm is proposed, to solve VRPTW with the objective of minimizing the total distance. In the BSO procedure, both inter-route and intra-route improvement heuristics are introduced. Experiments are conducted on Solomon's 56 instances with 100 customers benchmark, the results show that 42 out of 56 optimal solutions (18 best and 24 competitive solutions) are obtained, which illustrates the effectiveness of the proposed algorithm.

In this paper, the design of a synthetical absorber with topology optimization resistance frequency selective surface and metal wires array structure is proposed. This design attempts to provide a broadband strong absorption over wide-angle incidence at microwave band, especially for TE-polarization wave. Commonly, thickness is positively correlated with the absorption performance, and the thin thickness with high absorption of the absorber is taken into adequate consideration in our work. As a result, the optimal structure designed by genetic algorithm achieves absorption over a wide frequency range of 8.2 to 19.5GHZ with reflection below -10dB and the broadband absorption is tolerant for the incident angle under 70 degrees (high absorption less than 60 degrees). The results of experiment and simulation are highly consistent. What's more, the thickness of the resistive frequency selective surface and metal wires' dielectric substrate are 2mm and 0.4mm respectively, enabling a wide range of applications in stealth technology, electromagnetic radiation protection, electromagnetic control and so on.

Advances in the gene sequencing technology and the outbreak of artificial intelligence have made precision medicine a reality recently. Applying machine learning algorithms to cancer prediction using gene expression data helps to discover the link between genetic data and cancer, which will promote the development and application of precision medicine. Considering the natural order of genes, a new classification method that combines fused lasso and elastic net as regularization for linear support vector machine (SVM), which uses huberized hinge loss as the loss function, is proposed in this paper, which we name it oriented feature selection SVM (OFSSVM). Due to the characteristics of the elastic net and fused lasso, the OFSSVM can not only provide automatic feature selection, but also average the adjacent coefficients, resulting in a sparse and smooth solution. We demonstrate its effectiveness in both binary classification and multiclass classification in the sense of comprehensive evaluation that not only the classification accuracy but also the interpretability are considered. The experiments show that the OFSSVM is an appealing compromise between interpretability and classification accuracy, and is superior to other traditional methods in the sense of comprehensive evaluation.