LiDAR-based or RGB-D-based object detection is used in numerous applications, ranging from autonomous driving to robot vision. Voxel-based 3D convolutional networks have been used for some time to enhance the retention of information when processing point cloud LiDAR data. However, problems remain, including a slow inference speed and low orientation estimation performance. We therefore investigate an improved sparse convolution method for such networks, which significantly increases the speed of both training and inference. We also introduce a new form of angle loss regression to improve the orientation estimation performance and a new data augmentation approach that can enhance the convergence speed and performance. The proposed network produces state-of-the-art results on the KITTI 3D object detection benchmarks while maintaining a fast inference speed.
Due to the rapid development of chip technology and deep learning revolution, many ship detection frameworks for synthetic aperture radar (SAR) imagery based on convolutional neural networks (CNNs) have been proposed and achieved great success. However, there are problems hampering their development: 1) For the SAR ship detection task, it is uneconomic to apply heavy backbone network to extract features because it results in heavy computing load and prolongs the inference time cost; 2) The anchor-based methods usually have massive hyper-parameters, which typically need to be tuned carefully and easily lead to weak detection performance. To alleviate the problems, an efficient low-cost ship detection network for SAR imagery is proposed in this paper. Firstly, a simplified U-Net as the backbone to extract features is proposed. It only contains ∼ 0.47 million learnable weights, which is 2.37%, 0.76%, 0.34%, 1.01%, 0.55% and 1.07% of DarkNet-19, DarkNet-53, VGG-16, ResNet-50, ResNet-101 and ResNext-101, respectively. Secondly, an anchor-free SAR ship detection framework consisting of a bounding boxes regression sub-net and a score map regression sub-net based on simplified U-Net is proposed. To evaluate the effectiveness of our method, extensive experiments have been conducted and a more comprehensive set of evaluation metrics have been applied. Results demonstrate that the proposed network achieves 68.1% average precision and 67.6% average recall on the SAR ship detection dataset (SSDD), respectively. Compared with the state-of-the-art works, our proposed network achieves very competitive detection performance and extreme lightweight (∼ 0.93 million learnable weights in total).
Medical research shows that eye movement disorders are related to many kinds of neurological diseases. Eye movement characteristics can be used as biomarkers of Parkinson's disease, Alzheimer's disease (AD), schizophrenia, and other diseases. However, due to the unknown medical mechanism of some diseases, it is difficult to establish an intuitive correspondence between eye movement characteristics and diseases. In this paper, we propose a disease classification method based on decision tree and random forest (RF). First, a variety of experimental schemes are designed to obtain eye movement images, and information such as pupil position and area is extracted as original features. Second, with the original features as training samples, the long short-term memory (LSTM) network is used to build classifiers, and the classification results of the samples are regarded as the evolutionary features. After that, multiple decision trees are built according to the C4.5 rules based on the evolutionary features. Finally, a RF is constructed with these decision trees, and the results of disease classification are determined by voting. Experiments show that the RF method has good robustness and its classification accuracy is significantly better than the performance of previous classifiers. This study shows that the application of advanced artificial intelligence (AI) technology in the pathological analysis of eye movement has obvious advantages and good prospects.
Abstract:Limited by the properties of infrared detector and camera lens, infrared images are often detail missing and indistinct in vision. The spatial resolution needs to be improved to satisfy the requirements of practical application. Based on compressive sensing (CS) theory, this thesis presents a single image super-resolution reconstruction (SRR) method. With synthetically adopting image degradation model, difference operation-based sparse transformation method and orthogonal matching pursuit (OMP) algorithm, the image SRR problem is transformed into a sparse signal reconstruction issue in CS theory. In our work, the sparse transformation matrix is obtained through difference operation to image, and, the measurement matrix is achieved analytically from the imaging principle of infrared camera. Therefore, the time consumption can be decreased compared with the redundant dictionary obtained by sample training such as K-SVD. The experimental results show that our method can achieve favorable performance and good stability with low algorithm complexity.
In Wireless Sensor Networks (WSNs), unlicensed users, that is, sensor nodes, have excessively exploited the unlicensed radio spectrum. Through Cognitive Radio (CR), licensed radio spectra, which are owned by licensed users, can be partly or entirely shared with unlicensed users. This paper proposes a strategic bargaining spectrum-sharing scheme, considering a CR-based heterogeneous WSN (HWSN). The sensors of HWSNs are discrepant and exist in different wireless environments, which leads to various signal-to-noise ratios (SNRs) for the same or different licensed users. Unlicensed users bargain with licensed users regarding the spectrum price. In each round of bargaining, licensed users are allowed to adaptively adjust their spectrum price to the best for maximizing their profits. . Then, each unlicensed user makes their best response and informs licensed users of “bargaining” and “warning”. Through finite rounds of bargaining, this scheme can obtain a Nash bargaining solution (NBS), which makes all licensed and unlicensed users reach an agreement. The simulation results demonstrate that the proposed scheme can quickly find a NBS and all players in the game prefer to be honest. The proposed scheme outperforms existing schemes, within a certain range, in terms of fairness and trade success probability.
ObjectiveTo make a bibliometric analysis of global trends in research into exercise interventions for stroke between 2001 and 2021.MethodThis study did the systematic literature from 2001 to 2021 in Web of Science Core Collection. CiteSpace software was used to analyze the relationship of publications with countries, journals, authors, references, and keywords.ResultsA total of 3,484 publications were obtained in the bibliometric analysis. The number of publications increased gradually over the period. The United States have the most number of publications. The journal stroke had the most citations per paper (106.95) and the highest impact factor (IF 2020, 7.194). The most high frequency keywords are “stroke,” “rehabilitation,” and “recovery,” the top of burst key words are “health,” “speed,” and “aerobic exercise”.ConclusionThese findings provide the trends of exercise for stroke s and provided the potential research frontiers in the past 20 years. It will be a useful basis for further research into focus issues, cooperators, development trends.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
334 Leonard St
Brooklyn, NY 11211
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.