Deep learning with 3D data has progressed significantly since the introduction of convolutional neural networks that can handle point order ambiguity in cloud data. While being able to achieve good accuracies various scene understanding tasks, previous methods often have low training speed and complex network architecture. In this paper, we address these problems by proposing an efficient end-to-end permutation invariant convolution for deep learning. Our simple yet effective operator named...

Recent progresses in 3D deep learning has shown that it is possible to design special convolution operators consume point cloud data. However, a typical drawback rotation invariance often not guaranteed, resulting networks generalizes poorly arbitrary rotations. In this paper, we introduce novel operator for clouds achieves invariance. Our core idea use low-level invariant geometric features such as distances and angles learning. The well-known ordering problem also addressed by binning...

Self-attention based Transformer has demonstrated the state-of-the-art performances in a number of natural language processing tasks. is able to model long-term dependencies, but it may suffer from extraction irrelevant information context. To tackle problem, we propose novel called \textbf{Explicit Sparse Transformer}. Explicit improve concentration attention on global context through an explicit selection most relevant segments. Extensive experimental results series and computer vision...

Cross-domain (CD) hyperspectral image classification (HSIC) has been significantly boosted by methods employing Few-Shot Learning (FSL) based on CNNs or GCNs. Nevertheless, the majority of current approaches disregard prior information spectral coordinates with limited interpretability, leading to inadequate robustness and knowledge transfer. In this paper, we propose an asymmetric encoder-decoder architecture, Spectral Coordinate Transformer (SCFormer), for CDFSL HSIC task. Several dense...

Correlation analysis can reveal the complex relationships that often exist among variables in multivariate data. However, as number of grows, it be difficult to gain a good understanding correlation landscape and important intricate might missed. We previously introduced technique arranged into 2D layout, encoding their pairwise correlations. then used this layout network for interactive ordering axes parallel coordinate displays. Our current work expresses map employs visual analysis. In...

Skip connection is a widely-used technique to improve the performance and convergence of deep neural networks, which believed relieve difficulty in optimization due non-linearity by propagating linear component through network layers. However, from another point view, it can also be seen as modulating mechanism between input output, with scaled pre-defined value one. In this work, we investigate how scale factors effectiveness skip reveal that trivial adjustment will lead spurious gradient...

The fusion of hyperspectral imagery (HSI) and light detection ranging (LiDAR) data for classification has received widespread attention led to significant progress in research remote sensing applications. However, existing common CNN architectures suffer from the drawback not being able model images globally, while transformer are capture local features effectively. To address these bottlenecks, this paper proposes a framework multisource image fusion. First, spatial spectral feature...

Even though considerable progress has been made in deep learning-based 3D point cloud processing, how to obtain accurate correspondences for robust registration remains a major challenge because existing hard assignment methods cannot deal with outliers naturally. Alternatively, the soft matching-based have proposed learn matching probability rather than assignment. However, this paper, we prove that these an inherent ambiguity causing many deceptive correspondences. To address above...

Recent advances in deep learning for 3D point clouds have shown great promises scene understanding tasks thanks to the introduction of convolution operators consume directly a neural network. Point cloud data, however, could arbitrary rotations, especially those acquired from scanning. works show that it is possible design convolutions with rotation invariance property, but such methods generally do not perform as well translation-invariant only convolution. We found key reason compared...

Antibiotic contamination and its environmental impact in water-scarce human activity-intensive regions have been poorly researched, particularly the Zhuozhang River, China. Thus, this study investigated occurrence, sources, ecological risks of 27 different antibiotics based on water samples collected from representative locations including major reservoirs, upstream river, main river channel, a wastewater treatment plant (WWTP). Results showed widespread by quinolones, with concentrations...

High-resolution satellite images contain valuable road semantic information, but the occlusion of vegetation and buildings sparse distribution heterogeneous appearance roads limit accuracy extraction models. In this article, we propose a novel method for extracting using an ensemble learning model with postprocessing stage. The network weights biases our proposed deep are transmitted through random combination layers different submodels during forward backward propagation. gradient descent...

Abstract Imaging through scattering media is widely studied in various applications including industrial inspection and autonomous driving. Most existing image sensors always suffer from the aliasing effect that generates a diffused image, leading to an extremely challenging problem recovering targets of high quality, especially for 3D imaging. In this study, scanning‐driven photon‐counting imaging system proposed via asynchronous polarization modulation. To eliminate effect, we utilize...

An improved Vision Transformer (ViT) model incorporating the Canny algorithm is proposed to address low accuracy and efficiency issues in manually interpreting radiographic images for weld defect detection. The aim assist recognizing typical welding defects thereby enhance of interpretation. To model's capability capturing local information, an edge feature extraction module based on incorporated into original ViT architecture, leveraging advantage global perception while addressing its...