PCEDNet: A Lightweight Neural Network for Fast and Interactive Edge Detection in 3D Point Clouds

In recent years, Convolutional Neural Networks (CNN) have proven to be efficient analysis tools for processing point clouds, e.g., for reconstruction, segmentation, and classification. In this article, we focus on the classification of edges in point clouds, where both edges and their surrounding are described. We propose a new parameterization adding to each point a set of differential information on its surrounding shape reconstructed at different scales. These parameters, stored in a Scale-Space Matrix (SSM) , provide a well-suited information from which an adequate neural network can learn the description of edges and use it to efficiently detect them in acquired point clouds. After successfully applying a multi-scale CNN on SSMs for the efficient classification of edges and their neighborhood, we propose a new lightweight neural network architecture outperforming the CNN in learning time, processing time, and classification capabilities. Our architecture is compact, requires small learning sets, is very fast to train, and classifies millions of points in seconds.

Application of Image Mosaic Technology in Tai Chi Animation Creation

Panorama can reflect the image seen at any angle of view at a certain point of view. How to improve the quality of panorama stitching and use it as a data foundation in the “smart tourism” system has become a research hotspot in recent years. Image stitching means to use the overlapping area between the images to be stitched for registration and fusion to generate a new image with a wider viewing angle. This article takes the production of “Tai Chi” animation as an example to apply image stitching technology to the production of realistic 3D model textures to simplify the production of animation textures. A handheld camera is used to collect images in a certain overlapping area. After cylindrical projection, the Harris algorithm based on scale space is adopted to detect image feature points, the two-way normalized cross-correlation algorithm matches the feature points, and the algorithm to extract the threshold T iteratively removes mismatches. The transformation parameter model is quickly estimated through the improved RANSAC algorithm, and the spliced image is projected and transformed. The Szeliski grayscale fusion method directly calculates the grayscale average of the matching points to fuse the image, and finally, the best stitching method is used to eliminate the ghosting at the image mosaic. Data experiments based on Matlab show that the proposed image splicing technology has the advantages of high efficiency and clear spliced images and a more satisfactory panoramic image visual effect can be achieved.

Evaluating Scaling Frameworks for Multiscale Geomorphometric Analysis

Multiscale methods have become progressively valuable in geomorphometric analysis as data have become increasingly detailed. This paper evaluates the theoretical and empirical properties of several common scaling approaches in geomorphometry. Direct interpolation (DI), cubic convolution resampling (RES), mean aggregation (MA), local quadratic regression (LQR), and an efficiency optimized Gaussian scale-space implementation (fGSS) method were tested. The results showed that when manipulating resolution, the choice of interpolator had a negligible impact relative to the effects of manipulating scale. The LQR method was not ideal for rigorous multiscale analyses due to the inherently non-linear processing time of the algorithm and an increasingly poor fit with the surface. The fGSS method combined several desirable properties and was identified as an optimal scaling method for geomorphometric analysis. The results support the efficacy of Gaussian scale-space as a general scaling framework for geomorphometric analyses.

Research on Orbit Assembly Strategy of Large-scale Space Truss Structure

Background: With the rapid development of spatial technology and mankind's continuous exploration of the space domain, expandable space trusses play an important role in the construction of space station piggyback platforms. Therefore, the study of the in-orbit assembly strategy for space trusses has become increasingly important in recent years. The spatial truss assembly strategy proposed in this paper is fast and effective, and it is applied for the construction of future large-scale space facilities effectively. Objective: The four-prismatic truss periodic module is taken as the research object, and the assembly process of the truss and the assembly behaviors of the spatial cellular robot serving for on-orbit assembly are expressed. Methods: The article uses a reinforcement learning algorithm to study the coupling of truss assembly sequence and robot action sequence, then uses a q-learning algorithm to plan the strategy of the truss cycle module. Results: The robot is trained through the greedy strategy and avoids the failure problem caused by assembly uncertainty. The simulation experiment proves that the Q-learning algorithm of reinforcement learning used for planning the on-orbit assembly sequence of the truss periodic module structures is feasible, and the optimal assembly sequence with the least number of assembly steps obtained by this strategy. Conclusion: In order to address the on-orbit assembly issues of large spatial truss structures in the space environment, we trained the robots through greedy strategy to prevent failure due to the uncertainty conditions both in the strategy analysis and in the simulation study.Finally, the Q-learning algorithm in reinforcement learning is used to plan the on-orbit assembly sequence in the truss cycle module, which can obtain the optimal assembly sequence in the minimum number of assembly steps.

Scale-Covariant and Scale-Invariant Gaussian Derivative Networks

This paper presents a hybrid approach between scale-space theory and deep learning, where a deep learning architecture is constructed by coupling parameterized scale-space operations in cascade. By sharing the learnt parameters between multiple scale channels, and by using the transformation properties of the scale-space primitives under scaling transformations, the resulting network becomes provably scale covariant. By in addition performing max pooling over the multiple scale channels, or other permutation-invariant pooling over scales, a resulting network architecture for image classification also becomes provably scale invariant. We investigate the performance of such networks on the MNIST Large Scale dataset, which contains rescaled images from the original MNIST dataset over a factor of 4 concerning training data and over a factor of 16 concerning testing data. It is demonstrated that the resulting approach allows for scale generalization, enabling good performance for classifying patterns at scales not spanned by the training data.

Image Scale-Space Filtering Using Directional Local Variance Controlled Anisotropic Diffusion

The purpose of this paper is to develop an effective edge indicator and propose an image scale-space filter based on anisotropic diffusion equation for image denoising. We first develop an effective edge indicator named directional local variance (DLV) for detecting image features, which is anisotropic and robust and able to indicate the orientations of image features. We then combine two edge indicators (i.e., DLV and local spatial gradient) to formulate the desired image scale-space filter and incorporate the modulus of noise magnitude into the filter to trigger time-varying selective filtering. Moreover, we theoretically show that the proposed filter is robust to the outliers inherently. A series of experiments are conducted to demonstrate that the DLV metric is effective for detecting image features and the proposed filter yields promising results with higher quantitative indexes and better visual performance, which surpass those of some benchmark models.

APLICATION OF THE GAUSSIAN MODEL IN RESEARCH SPREADING AIR POLLUTION

The article deals with the application of the Gaussian distribution for calculating the concen-tration of pollutants. When using the principle of superposition, we have the opportunity to obtain models for calculating the concentration of impurities from a point source of continuous action, in-stantaneous areal and instantaneous volumetric sources. The obtained standard deviations make it possible to assess the effect of air turbulence on the dispersion of pollutants. The first Gaussian model allows one to obtain a diffusion model of a local small-scale space and make predictions, then, based on the Gaussian model of the study, a modified model is obtained for other reliefs and weather conditions. Therefore, the modeling accuracy and applicable conditions are difficult to cope with the needs of large-scale complex meteorological conditions of air quality models.

A boundary division guiding synchrosqueezed wave packet transform method for rolling bearing fault diagnosis

Synchrosqueezed wave packet transform (SSWPT) can effectively reconstruct the band-limited components of the signal by inputting the specific reconstructed boundaries and it provides an alternative bearing fault diagnosis method. However, the selection of reconstructed boundaries can significantly affect the fault feature extraction performance of SSWPT. Accordingly, this paper presents a boundary division guiding SSWPT (BD-SSWPT) method. In this method, an adaptive boundary division method is developed to effectively determine the reconstructed boundaries of SSWPT. Firstly, the marginal spectrum of SSWPT, more robust to noise than the Fourier spectrum, is defined for the scale-space division to obtain the initial boundaries. Secondly, the inverse transform of SSWPT is conducted based on the initial boundaries to obtain the initial reconstructed components. Thirdly, a boundary redefinition scheme, composed of clustering and combination, is conducted to redefine the boundaries. Finally, the potential components are extracted by the inverse transform of SSWPT based on the redefined boundaries. The validity of BD-SSWPT is verified by simulated and experimental analysis, and the superiority of BD-SSWPT is highlighted through comparison with singular spectrum decomposition (SSD) and an adaptive parameter optimized variational mode decomposition (AVMD). The results demonstrate that BD-SSWPT identifies more significant fault features and has higher computational efficiency than SSD and AVMD.

Scale-space energy density for inhomogeneous turbulence based on filtered velocities

The energy spectrum is commonly used to describe the scale dependence of turbulent fluctuations in homogeneous isotropic turbulence. In contrast, one-point statistical quantities, such as the turbulent kinetic energy, are mainly employed for inhomogeneous turbulence models. Attempts have been made to describe the scale dependence of inhomogeneous turbulence using the second-order structure function and two-point velocity correlation. However, unlike the energy spectrum, expressions for the energy density in the scale space fail to satisfy the requirement of being non-negative. In this study, a new expression for the scale-space energy density based on filtered velocities is proposed to clarify the reasons behind the negative values of the energy density and to obtain a better understanding of inhomogeneous turbulence. The new expression consists of homogeneous and inhomogeneous parts; the former is always non-negative, while the latter can be negative because of the turbulence inhomogeneity. Direct numerical simulation data of homogeneous isotropic turbulence and a turbulent channel flow are used to evaluate the two parts of the energy density and turbulent energy. It was found that the inhomogeneous part of the turbulent energy shows non-zero values near the wall and at the centre of a channel flow. In particular, the inhomogeneous part of the energy density changes its sign depending on the scale. A concave profile of the filtered-velocity variance at the wall accounts for the negative value of the energy density in the region very close to the wall.